Skip to main content

Full text of "Chace The Economic Singularity; Artificial Intelligence And The Death Of Capitalism ( 2016)"

See other formats





Artificial intelligence and the death of capitalism 

"Read The Economic Singularity if 
you want to think intelligently about 
the future." Aubrey de Grey 

The Economic Singularity 

Artificial intelligence and the death of 


by Calum Chace 

Table of Contents 

Chapter 1. Introduction: the economic singularity 8 

Chapter 2. The History of Automation 11 

2.1- The industrial revolution 11 

2.2 - The information revolution 13 

2.3 - The Automation story so far 15 

2.3 - The Automation story so far 15 

2.4 - The Luddite fallacy 20 

3 - Is it different this time? 23 

3.1- Prophets of change 24 

3.2 - Academic and consultancy studies 30 

3.3 - Crying wolf 34 

3.4-Alto date 37 

3.5 - Exponential future 46 

3.6 - What people do 52 

3.7- Related technologies 55 

3.8 - The poster child for technological unemployment: self-driving vehicles 67 

3.9 - Who's next? 74 

3.10 - Jobs or no jobs 84 

3.11- What's the problem? 93 

3.12- Conclusion: yes, it’s different this time 95 

4. - A timeline 97 

4.1- Un-forecasts 97 

4.2- 2021 100 

4.3- 2031 102 

4.4- 2041 105 

5. - The Challenges 108 

5.1 - Economic contraction 109 

5.2- Distribution 110 

5.3- Meaning 118 

5.4 - Allocation 121 

5.5- Cohesion 124 

6. - Scenarios 128 

6.1 - No Change 128 

6.2 - Racing with the machines 130 

6.3 - Capitalism + UBI 132 

6.4- Fracture 133 

6.5- Collapse 134 

6.6- Protopia 135 

Chapter 7. Summary and recommendations 139 

7.1- The argument 139 

7.2 - The two singularities 143 

7.3 - What is to be done? 144 

Acknowledgements 148 

Comments on The Economic Singularity 

A problem that all techno-pioneers face, when “selling” their vision of the future to 
others (I use quotes because I in no way refer to anything monetary), is to get their 
audience to focus on the new development in an appropriate context. Above all, this 
means striking a balance between communicating the significance of the proposed 
development and setting it within the universe of other developments that are likely to 
have occurred in the meantime. 

The advance of automation, described with great care and accuracy in this book, will 
almost certainly constitute the substrate within which all other technological 
developments - be they biomedical, environmental or something else entirely - will 
occur, and thus within which they should be discussed as regards their value to 

Read "The Economic Singularity" if you want to think intelligently about the future. 
Aubrey de Grey - CSO of SENS Research Foundation; former AI researcher 

Following his insightful foray into the burgeoning AI revolution and associated 
existential risks, Calum focuses his attention on a nearer term challenge - the likelihood 
that intelligent machines will render much of humanity unemployable in the foreseeable 
future. He explores the arguments for and against this assertion and provides a 
measured response, acknowledging the risks associated with such a radical shift in our 
self identity but also outlining the potential significant benefits. Once again he proves a 
reliable guide through this complex yet fascinating topic. 

Ben Medlock, co-founder of Swiftkey, the best-selling app on Android 

"It's important that this book and others like it are written. Not because the future 
will necessarily happen exactly in the way described, but because it's important to 
be prepared if it does. If automation compels us to shift to a different economic 
organisation, we better start laying the foundations for the shift right now." 

Dr Stuart Armstrong, James Martin Research Fellow at the Future of 

Humanity Institute, Oxford University 

"Chace does a good job of answering the question whether robots will take our 
jobs. What worries me more though, a bit further down the road, once these 
robots have become massively intelligent, is whether they may take our lives. 

Chace covered this issue thoroughly in his previous book, "Surviving AI". 

Prof. Dr. Hugo de Garis - author of "The Artilect War", former director of the 
Artificial Brain Lab, Xiamen University, China 

“The jobs of the future don’t exist today and the jobs of today will not exist in the future. 
Technological Singularity will change everything, but its first manifestation will come 
in the domain of economics, most likely in the shape of technological unemployment. 
Calum Chace’s “The Economic Singularity” does a great job of introducing readers of 
all levels to the future we are about to face. Chace explains what might happen and 
what we can do to mitigate some of the negative consequences of machine takeover. The 
book covers unconditional basic income, virtual environments, and alternative types of 
economies among other things. Highly recommended.” 

Dr. Roman V. Yampolskiy, Professor of Computer Engineering and Computer 
Science, Director of Cybersecurity lab, Author of Artificial Superintelligence: a 
Futuristic Approach 

Unprecedented productivity gains and unlimited leisure—what could possibly go 
wrong? Everything, says Calum Chace, if we don’t evolve a social system suited to 
the inevitable world of connected intelligent systems. 

It’s a failure of imagination to debate whether there will be jobs for humans in the 
automated world, Chace argues - we must look farther and ask how we will organize 
society when labor is not necessary to provide for the necessities of life. Find an 
answer, and life improves for all; without one, society collapses. Read this book to 
understand how social and technological forces will conspire to change the world—and 
the problems we need to solve to achieve the promise of the Economic Singularity. 

Christopher Meyer, author of “Blur”, “Future Wealth”, and “Standing on the 

It is interesting to listen to our own language. We say things such as "to earn a living", 
implying that you need to earn the privilege to be alive and to live a moderately 
enjoyable life. This may be looked upon as strangely in the future as we would now 
look back and say about slaves that they had to "earn their freedom". 

Calum Chace hits the nail on the head in chapter 3 of this extremely timely book. It is 
probably true that there will be new types of'jobs' in whatever niches remain best 
explored by humans in the near future, but we should also consider an entirely different 
goal for the future. 

Who was it in ages past who contributed those things we most remember, over time, as 
being of great value? It was they who contributed to the arts, the sciences and 
invention. But who were those people? Throughout the majority of history, these were 
mainly the people who either did not have to have a 'job' (because they were part of an 
aristocracy that had a different role to play while being supported by property and 
subjects), as well as the artists, artisans, philosophers or scientists who were directly 
supported by those patrons and therefore did not have the need to take a typical 'job'. 

It is not the typical jobs that are celebrated as the best of humanity, and therefore it 
probably should not be our aim to find yet more categories of such jobs. Instead, 
wouldn't it be much better if a greater proportion of humanity could find the means to 
engage in preferred and culture-creating activity? With this in mind, it seems to me that 
it should be our aim to get rid of the need for jobs and employment just for the purpose 
of survival. 

Our strategies for the future should be not about finding new salary jobs, but rather 
about removing the need for them, and about setting up a better and more advanced 
social structure. This is where looking at the challenges involved and the path to a 
successful alternative, as Chace does in chapter 5, is essential. 

Where ideas such as a universal basic income (UBI) are concerned, it is useful to keep 
in mind that the world is not the US. Even if there is some initial antipathy in the US, 
because of associations between UBI and what might naively be labeled as 'socialist' 
thinking, the US will not wish to be left behind if other nations successfully implement 
the change. The time to dive deeply into the many issues raise in this book, to start a 
wider conversation about those issues, and to look creatively for the most well- 
balanced solutions and outcomes, is now. 

Randal Koene -founder of 

"The Economic Singularity is fascinating. Calum Chace brilliantly explores the 
enormous opportunities, and risks, presented to humanity by the rapid advance of 
technology, and especially artificial intelligence. I couldn't put this book down." 

Ben Goldsmith - Menhaden Capital 

In his fast-paced new book, Calum Chace explains the challenge facing humanity: to 
navigate through a dramatic transition which he christens the economic singularity. The 
culmination of an accelerating wave of automation by robots and AI, this transition 
threatens to do more than displace employees from the workforce. Unexpectedly, it 
threatens the end of capitalism itself, and potentially the fracturing of the human 

Chace compellingly sets out a range of options, before sharing his assessment of the 
most credible and desirable outcomes, so that we can reach a shared “protopia” rather 
than a nightmarish “Brave New World” (or worse). 

David Wood - chairman, London Futurists 

Calum Chace is a best-selling author of fiction and non-fiction books, focusing on the 
subject of artificial intelligence. His books include “Surviving AI”, a non-fiction book 
about the promise and the challenges of AI, and “Pandora's Brain”, a techno-thriller 
about the first superintelligence. 

He is a regular speaker on artificial intelligence and related technologies and runs a 
blog on the subject at . 

Before becoming a full-time writer, Calum had a 30-year career in journalism and 
business, in which he was a marketer, a strategy consultant and a CEO. He maintains 
his interest in business by serving as chairman and coach for a selection of growing 
companies. In 2000 he co-wrote “The Internet Startup Bible”, a business best-seller 
published by Random House. 

A long time ago, Calum studied philosophy at Oxford University, where he discovered 
that the science fiction he had been reading since boyhood is actually philosophy in 
fancy dress. 

Also by Calum Chace 
Surviving AI 
Pandora’s Brain 

The Internet Startup Bible (co-authored) 
The Internet Consumer Bible (co-authored) 

For Julia and Alex 


A Three Cs book 
ISBN 978-0-993-21164-5 

First published in 2016 by Three Cs 
Copyright © CalumChace 2016 

Cover and interior design © Rachel Lawston at 
Lawston Design, 
Photography © and 

All rights reserved 

The right of Calum Chace to be regarded as the author of this work has been asserted by 
him in accordance with the Copyright, Design and Patents Act 1988 

Chapter 1. Introduction: the economic singularity 

Accelerating change 

In the next few decades, life for most people is going to change in extraordinary ways 
and at an extraordinary rate. The reason, as usual, is technology. 

Human lives and societies can be transformed by religions, cultural memesJTl, an d the 
imposition of new economic systems. They can be transformed by the passion and 
belief of a single great man or woman. But when profound and lasting change takes 
place it is usually because we found a new way of doing things - a new technology. 

Thus we name many historical periods after their dominant technology: the iron age, the 
bronze age, and so on. 

When a cluster of related technological innovations come along together they can create 
sufficient change to merit the title of a revolution. This has happened twice before in 
human history, with the agricultural and the industrial revolutions, and we are now in 
the middle of a third, the information revolution. 

These revolutions are not overnight affairs: the industrial revolution has been under way 
for 300 years. [fij The information revolution is just half a century old, [in] and in many 
ways we are nearer to the start than the end. We think the world has changed greatly in 
the last century, and especially in the last twenty years or so, and indeed it has. But the 
rate of change is accelerating, and the changes that are coming will dwarf what has 
happened so far. 

Forecasting has always been perilous. Throughout history, most long-term forecasts 
have been wrong, often blind-sided by the arrival of a new technology like 
smartphones. But in the coming decades the rate and scale of change will be so great 
that the future will become mysterious in a new way. So much so that people talk about 
a coming technological singularity. 

The term “singularity” is borrowed from maths and physics, where it means a point at 
which a variable becomes infinite. The usual example is the centre of a black hole, 
where matter becomes infinitely dense. When you reach a singularity, the normal rules 
break down, and the future becomes even harder to predict than usual. In recent years, 

the term has been applied to the impact of technology on human affairs, [iv] 

Superintelligence and the technological singularity 

The technological singularity is most commonly defined as what happens when the first 
artificial general intelligence (AGI) is created - a machine which can perform any 
intellectual task that an adult human can. It continues to improve its capabilities and 
becomes a superintelligence, much smarter than any human. It then introduces change to 
this planet on a scale and at a speed which un-augmented humans cannot comprehend. I 
wrote about this extensively in my book, “Surviving Af\ 

The term “singularity” became associated with a naive belief that technology, and 
specifically a superintendent AI, would magically solve all our problems, and that 
everyone would live happily ever after. Because of these quasi-religious overtones, the 
singularity was frequently satirised as “rapture for nerds”, and many people felt 
awkward about using the term. 

The publication in 2014 of Nick Bostrom's seminal book “Superintelligence” was a 
watershed moment, causing influential people like Stephen Hawking, Elon Musk and 
Bill Gates to speak out about the enormous impact which AGI will have - for good or 
for ill. They introduced the idea of the singularity to a much wider audience, and made 
it harder for people to retain a blinkered optimism about the impact of AGI. 

For time-starved journalists, “good news is no news” and “if it bleeds it leads”, so the 
comments of Hawking and the others were widely mis-represented as pure doom¬ 
saying, and almost every article about AI carried a picture of the Terminator. AI 
researchers and others hastened to warn us (rightly) not to throw the baby of AI out with 
the bathwater of unfriendly superintelligence, and the debate is now more nuanced. 

Technological unemployment and the economic singularity 

So for me at least, the term “singularity” no longer seems so awkward. And it seems 
reasonable to apply it to another event which is likely to take place well before the 
technological singularity. I call this event “the economic singularity”. 

There is a lot of talk in the media at the moment about technological unemployment - the 
process of people becoming unemployed because machines can do any job that they 
could do, and do it cheaper, faster and better. There is widespread disagreement about 
whether this is happening already, whether it will happen in the future, and whether it is 

a good or a bad thing. This disagreement is natural and inevitable: one of the main 
features of a singularity is that what lies beyond its event horizon is hard to see[v]. 
Nevertheless we must try to peer into the hazy future if we are to prepare ourselves for 

In this book I will argue that technological unemployment is not happening yet (or at 
least, not much), that it will happen in the next few decades, and that it can be a very 
good thing indeed - if we prepare for it, and manage the transition successfully. 

Naturally, there are challenges. As we will see, a lot of people believe that Universal 
Basic Income (UBI) is a silver bullet that will solve the problem of technological 
unemployment. UBI is a guaranteed income paid to all citizens simply because they are 
citizens. It may take some time for the idea of UBI to be accepted, especially in the 
USA, where resistance to anything that smacks of socialism is often fierce - almost 
visceral. Martin Ford's otherwise excellent book “The Rise of the Robots” almost 
fizzles out at the end because he seems daunted by the scale of the opposition that UBI 
will face in his home country. 

But to my mind, UBI is not the real battle. In Europe we are very comfortable with the 
idea of a safety net of welfare programmes which prevent the economically 
unsuccessful from falling into absolute penury. Most American states provide 
unemployment benefits too, although they usually cease after six months. In fact the US 
spends more per capita on welfare ($650 in 2011, according to the OECD) than the UK 
($610) or Canada t$550T[vi] Unlike some of my American friends, I believe the 
people of that great country will quickly accept the need for UBI if and when it becomes 
undeniable that the majority of them are going to be unemployable. 

The real problem, it seems to me, is that we will need more than just UBI. We may 
need an entirely new form of economy. I see great danger in a world in which most 
people rub along on handouts while a minority - perhaps a tiny minority - not only own 
most of the wealth (that is pretty much true already) but are the only ones actively 
engaged in any kind of economic activity. Given the advances in all kinds of technology 
that we can expect in the coming decades, this minority would be under immense 
temptation to separate themselves off from the rest of us - not just economically, but 
cognitively and physically too. Yuval Harari, author of the brilliant book “Sapiens”, 
says that in the coming century or so, humanity will divide into two classes of people. 
Rather brutally, he calls them the gods and the useless. [vii] 

Capitalism and liberal democracy have served humanity well in the last couple of 
centuries. I am not convinced they will continue to do so in a post-automation world, 

but it is no small task to work out what they should be replaced with, and how that can 
be achieved without turmoil. This sounds like a singularity - an economic singularity. 

Chapter 2. The History of Automation 
2.1- The industrial revolution 

For a process that began hundreds of years ago, the start date for the industrial 
revolution is surprisingly controversial. Historians and economists cannot even agree 
how many industrial revolutions there have been: some say there has been one 
revolution with several phases, others say there have been two, and others say more. 

The essence of the industrial revolution was the shift from manufacturing goods by hand 
to manufacturing them by machine, and the harnessing of better power sources than 
animal muscle. So a good date for its beginning is 1712, when Thomas Newcomen 
created the first practical steam engine for pumping water. For the first time in history, 
humans could generate more power than muscles could provide - wherever they needed 

The replacement of human labour by machines in manufacturing dates back considerably 
earlier, but they were powered by muscles or by wind or water. In the 15 th century, 
Dutch workers attacked textile looms by throwing wooden shoes into them The shoes 
were called sabots, and this may be the etymology of the word “saboteur”. A century 
later, around 1590, Queen Elizabeth (the First) of England refused a patent to William 
Lee for a mechanical knitting machine because it would deprive her subjects of 

In the second half of the 18 th century, the Scottish inventor James Watt teamed up with 
the English entrepreneur Matthew Boulton to improve Newcomen’s steam engine so that 
it could power factories, and make manufacturing possible on an industrial scale. At the 
same time, iron production was being transformed by the replacement of charcoal by 
coal, and “canal mania” took hold, as heavy loads could be transported more cheaply by 
canal than by road or sea. 

Later, in the mid-19 th century, steam engines were improved sufficiently to make them 
mobile, which ushered in the UK's “railway mania” of the 1840s. Projects authorised in 
the middle years of that decade led to the construction of 6,000 miles of railway - more 
than half the length of the country's current rail network. Other European countries and 
the USA emulated the UK's example, usually lagging it by a decade or two. 

Toward the end of the 19 th century, Sir Henry Bessemer's method for converting iron 
into steel enabled steel to replace iron in a wide range of applications. Previously, 

steel had been an expensive commodity, reserved for specialist uses. The availability 
of affordable steel enabled the creation of heavy industries, building vehicles for road, 
rail, sea and later the air. 

As the 20 th century arrived, oil and electricity provided versatile new forms of power 
and the industrial world we recognise today was born. The changes brought about by 
these technologies are still in progress. 

In summary, we can identify four phases of the industrial revolution: 

1712 onwards: the age of primitive steam engines, textile manufacturing machines, and 
the canals 

1830 onwards: the age of mobile steam engines and the railways 

1875 onwards: the age of steel and heavy engineering, and the birth of the chemicals 


1910 onwards: the age of oil, electricity, mass production, cars, planes and mass travel. 

From an early 2 l st -century standpoint, it seems entirely natural that the industrial 
revolution took off where and when it did. In fact it is something of a mystery. Western 
Europe was not the richest or most advanced region of the world: there were more 
powerful empires in China, India and elsewhere. There is still room for debate about 
whether the technological innovations came about in England at that time because of the 
cultural environment, the legal framework, or the country's fortuitous natural resources. 
Fascinating as these questions are, they need not detain us. 

2.2 - The information revolution 

Even though the industrial revolution is still an on-going process, there is general 
agreement that we are now in the process of an information revolution. There is less 
consensus over when it began or how long it is likely to continue. 

The distinguishing feature of the information revolution is that information and 
knowledge became increasingly important factors of production, alongside capital, 
labour, and raw materials. Information acquired economic value in its own right. 
Services became the mainstay of the overall economy, pushing manufacturing into 
second place, and agriculture into third. 

One of the first people to think and write about the information revolution and the 
information society was Fritz Machlup, an Austrian economist. In his 1962 book, The 
production and distribution of knowledge in the United States , he introduced the 
notion of the knowledge industry, by which he meant education, research and 
development, mass media, information technologies, and information services. He 
calculated that in 1959, it accounted for almost a third of US GDP, which he felt 
qualified the US as an information society. 

Alvin Toffler, author of the visionary books Future Shock { 1970) and The Third Wave 
(1980), argued that the post-industrial society has arrived when the majority of workers 
are doing brain work rather than personally manipulating physical resources - in other 
words when they are part of the service sector. Services grew to 50% of US GDP 
shortly before 1 940. [viii] and they first employed the majority of working Americans 
around 1950. 

We have seen that the start and end dates of the economic revolutions (agricultural, 
industrial and information) are unclear. What's more, they can overlap, and sometimes 
re-ignite each other. 

An example of this overlap is provided by the buccaneers who preyed on Spanish 
merchant shipping en route to and from Spain's colonies in South America during the 
17 th century. (Some of these buccaneers were effectively licensed in their activities by 
the English, French and Dutch crowns, which issued them with “letters of marque”. 

This ceased when Spain's power declined toward the end of the century, and the 
buccaneers became more of a nuisance than a blessing to their former sponsors.) When 
a buccaneer raiding party boarded a Spanish ship the first thing they would look for and 

demand was the maps. Charts - a form of information which improve navigation - 
were actually more valuable than silver and gold.[ix] 

An example of one revolution re-igniting another is that the industrial revolution 
enabled the mechanisation of agriculture, causing a second agricultural revolution, 
making the profession of farming more effective and more efficient. The information 
revolution does the same, providing farmers with crops that are more resilient in the 
face of weather, pests and weeds, and allowing them to sow, cultivate and harvest their 
crops far more accurately with satellite navigation. 

Along with the uncertainty about the start date of the information revolution, there is 
disagreement about how distinct it is from the industrial revolution. The Internet of 
Things (IoT) is a phenomenon of the information revolution which we will look at in 
more detail in chapter 3.7. Klaus Schwab, founder and executive chairman of the 
World Economic Forum which hosts the annual meeting of the global elite in Davos, 
calls the IoT the fourth industrial revolution.^ This seems to me to under-state the 
importance of the IoT, and also to separate it from all the other digital revolutions 
which comprise the information revolution, including, of course, artificial intelligence. 

2.3 - The Automation story so far 
The mechanisation of agriculture 

The particular aspect of the industrial and information revolutions which concerns us in 
this book is automation. Perhaps the clearest example of automation destroying jobs is 
the mechanisation of agriculture, a sector which accounted for 41% of US employment 
in 1900, and only 4% by 1970[xi], (The corresponding figures for the UK are lower in 
absolute terms, but similar in relative terms: 9% in 1900 falling to 1% in2000. [xii] l 

Many of the people who quit farm work moved to towns and cities to take up other jobs 
because they were easier, safer, or better paid. Many others were forced to find 
alternative employment because they could not compete with the machines. This 
process caused much suffering to individuals, but overall, the level of employment did 
not fall, and society became far richer - both in total and on average. More than one 
new job was created for every job that was lost. 

The reason for this is that as machines replaced muscle power on the farm, humans had 
other skills and abilities to offer. Factories and warehouses took advantage of our 
manual dexterity and our ability to carry out a very broad range of activities. Office 
jobs used our cognitive ability. We turned our hands (often literally) to more value¬ 
adding work: you could say that we climbed higher up the value chain. 

One-trick ponies 

While the mechanisation of agriculture was a good news story for humans, it was less 
positive for the horse, which had nothing to offer beyond muscle power. 1900 was 
probably when the US reached “peak horse”, with a population of 2 lm. That number 
fell to just 3m by 1960 [xiii] . 

Artificial intelligence systems and their peripherals, the robots, are increasingly 
bringing flexibility, manual dexterity, and cognitive ability to the automation process. 
One of the big questions addressed in this book is: as computers take over the role of 
ingesting, processing and transmitting information, will there be anywhere higher up the 
value chain for humans to retreat to? In other words, can we avoid playing the role of 
the horse in the next wave of automation? Are we approaching “peak human” in the 

Mechanisation and automation 

What went on in farms was mechanisation rather than automation, and the distinction is 
important. Mechanisation is the replacement of human and animal muscle power by 
machine power; a human may well continue to control the whole operation. Automation 
means that machines are controlling and overseeing the process as well: they 
continuously compare the operation to a pre-set set of parameters, and adjust the 
process if necessary. 

Although the word "automation" was not coined until the 1940s by General 
Electric, [xiv] this description applies pretty well to the operation of 19 th -century steam 
engines once James Watt had perfected his invention of governors. Automated 
controllers which were able to modify the operation more flexibly became increasingly 
common in the early 20 th century, but the start-stop decisions were still normally made 
by humans. 

In 1968 the first programmable logic controllers (PLCs) were introduced[xv]. These 
are rudimentary digital computers which allow far more flexibility in the way an 
electrochemical process operates, and eventually general-purpose computers were 
applied to the job. 

The advantages of process automation are clear: it can make an operation faster, 
cheaper, and more consistent, and it can raise quality. The disadvantages are the initial 
investment, which can be substantial, and the fact that close supervision is often 
necessary. Paradoxically, the more efficient an automated system becomes, the more 
crucial the contribution of the human operators. If an automated system falls into error it 
can waste an enormous amount of resources and perhaps cause significant damage 
before it is shut down. 

Let's take a look at how automation has affected some of the largest sectors of the 

Retail and “prosumers” 

Retail is a complicated business and there have been attempts to automate many of the 
processes required to get goods from supplier to customer, and payment from customer 
to supplier. Demand forecasting, product mix planning, purchasing, storage, goods 
handling, distribution, shelf stacking, customer service and many other aspects of the 

business have been automated to varying extents in different places and at different 

The retail industry has also given us the clearest examples of another, associated 
phenomenon known as “prosumption”. This term was coined in 1980 by the American 
futurist Alvin Toffler, one of the leading thinkers about the trends we are discussing in 
this book. At the same time as organisations automate many of their processes, they 
enlist the help of their customers to streamline their operations. In fact, they get their 
customers to do some of the work that was previously done for them. The reason why 
consumers accept this (indeed welcome it) is that the process speeds up, and becomes 
more flexible - more tailored to their wishes. 

Toffler first described this process in FutureShock (1970), and in The Third Wave 
(1980) he defined a “prosumer” as a consumer who is also involved in the production 
process. Where once people were passive recipients of a limited range of goods and 
services designed or selected by retailers, he foresaw that we would become 
increasingly involved in their selection and configuration. 

Perhaps the simplest example of what he meant is the purchase of gasoline. This 
dangerous substance was traditionally dispensed by pump attendants, but Richard 
Corson’s invention of the automatic shut-off valve enabled the job to be taken over by 
customers. Nowadays most consumers in developed countries dispense their own 
gasoline at self-service pumps. This saves money for the retailer and time for the 
consumer, [xvi] 

Supermarkets have often led the way in automation and prosumption because they are 
owned by massive organisations with the budgets and the sophistication to invest in the 
systems needed. Once upon a time, what marketers call fast-moving consumer goods 
(foods, toiletries, etc.) were requested one at a time by the shopper at a counter and 
fetched individually by the shopkeeper or his assistant. As these general stores firms 
grew bigger and more sophisticated they built large stores where shoppers fetched their 
own items, and presented them for processing at checkouts, like components on a car 
assembly line. Later on, self-service tills were installed, where shoppers could scan 
the bar codes of their goods themselves, speeding up the process considerably. Soon, 
RFID tags jxvii] on goods will enable you to wheel your trolley lull of items out of the 
store and to your car without the fuss of unloading and re-loading them at a checkout. 

At each stage of this evolution, the involvement of the consumer in selecting and 
transporting each item increases, and the requirement for shop staff involvement 
reduces. This latter effect is disguised because, as society gets richer, people buy many 

more items, so the store needs more staff even though their involvement in each 
individual item is less. 

Online shopping is perhaps the ultimate prosumer experience. Consumer reviews 
replace the retailer’s sales force, and its algorithms do the up-selling. 

Call centres 

Of course, automation and prosumption is not always to the benefit of consumers. In 
markets where switching costs or partial monopolies dilute the standards-raising effect 
of competition, companies can save money for themselves in ways which actually make 
life worse for their customers. We are all familiar with call centres where (for 
instance) utility companies and banks have automated their customer service operations, 
obliging frustrated customers to plough through various levels of artificial un¬ 
intelligence in order to get their problem resolved. The customer would be much better 
off if a human picked up the call immediately, but that would cost the companies a lot 
more money, and they have no incentive to incur that cost. 

Things are improving, however, as the AI used in call centres advances. Just as most 
people choose to withdraw cash from ATMs rather than venture into the bank and wait 
in line for a human cashier, many call centre operations are now getting good enough at 
handling or triaging problems that we may soon prefer to deal with the automated 
system than with a human. 

Food service 

The automation of service in fast food outlets seems to have been just around the corner 
for decades. Indeed, elements of it have been a reality for years in Oriental-style outlets 
like Yo, Sushi!, but it has so far failed to spread to the rest of the sector. There are 
several reasons for this, including the relatively low labour cost of people working in 
fast food outlets, and the need for every single purchase to be problem-free, and if not, 
for there to be a trained human on hand to solve any problem immediately. If a hands¬ 
free wash basin fails 5% of the time it is no big deal, but it would be a very big problem 
if 5% of meals were inedible, or delivered to the wrong customer. Three restaurants in 
Guangzhou, southern China, which trumpeted their use of robot waiters had to abandon 
the practice because the machines simply weren’t good enough. [xviii] 

A combination of factors is poised to overcome this resistance. Increases to the cost of 
labour caused by rising minimum wage legislation, declining costs of the automated 

technology, greater cultural acceptance of interacting with machines, and above all, the 
improved performance of the automated technology. It is increasingly flexible, and it 
goes wrong less often. McDonald's is one of the many fast food chains that are 
introducing touch-screen ordering and payment systems in their restaurants, and it is 
trialling an automated McCafe kiosk in a restaurant in Chicago. [xix] KFC (formerly 
known as Kentucky Fried Chicken) has a store in Shanghai where customers’ orders are 
taken by a robot equipped with voice recognition software. [xx] Our robot overlords 
have found a Colonel. 


Car manufacturing has traditionally incurred relatively high labour costs. The work 
involves a certain amount of physical danger, with heavy components being transported, 
and metals being cut and welded. It is also a sector where a lot of the operations can be 
precisely specified and were highly repetitive. These characteristics make it ripe for 
automation, and the fact that the output (cars) are high-value items means that investment 
in expensive automation systems can be justified. Around half of all the industrial 
robots in service today are engaged in car manufacturing, [xxi] 

Despite the recession, sales of industrial robots grew at 10% a year from 2008 to 2013, 
when 178,000 were sold worldwide. Sales in 2014 jumped 29% to 229,000, and the 
International Federation of Robotics expects the number to jump a further 75% to 
400,000 by 2018. China became the biggest market in 2013, installing 37,000 robots 
compared with 30,000 in the USA [xxii] , 

Until recently, the industrial robots used in car manufacturing (and elsewhere) were 
expensive, inflexible, and dangerous to be around. But the industrial robotics industry 
is changing: as well as growing quickly, its output is getting cheaper, safer and far more 

A landmark was reached in 2012 with the introduction of Baxter, a 3-foot tall robot (6 
feet with his pedestal) from Rethink Robotics. The brainchild of Rodney Brooks, an 
Australian roboticist who used to be the director of the MIT Computer Science and 
Artificial Intelligence Laboratory, Baxter is much less dangerous to be around. By early 
2015, Rethink had received over $ 100m in funding from venture capitalists, including 
the investment vehicle of Amazon founder Jeff Bezos. Baxter was intended to disrupt 
the industrial robots market by being cheaper, safer, and easier to programme. He is 
certainly cheaper, with a starting price of $22,000. He is safer because his arm and 
body movements are mediated by springs, and he carries an array of sensors to detect 

the presence nearby of squishy, fragile things like humans. He is easier to programme 
because an operator can teach him new movements simply by physically moving his 
arms in the intended fashion. 

Baxter's short life has not been entirely plain sailing. Sales did not pick up as expected, 
and in December 2013, Rethink laid off around a quarter of its staff. One of the 
competitors stealing sales from Rethink is Universal Robots of Denmark, a manufacturer 
of small- and medium-sized robot arms. Universal increased sales to €30m in 2014, 
and aims to double its revenues every year until 2017. 

But Rethink remains well-funded, and in March 2015 it introduced a smaller, faster, 
more flexible robot arm called Sawyer. It can operate in more environments than 
Baxter, and can carry out more intricate movements. It is slightly more expensive, at 
$29,000. Rethink and Universal, along with other companies like the Swiss firm ABB 
and the German firm KUKI, are making industrial robots more effective, more 
affordable, and more widespread. 


Kiva Systems was established in 2003, and acquired by Amazon in 2012. Kiva 
produces robots which collect goods on pallets from designated warehouse shelves and 
deliver them to human packers in the bay area of the warehouse. Amazon paid $775m 
for the nine-year old company and promptly dispensed with the services of its sales 
team. Re-named Amazon Robotics in August 2015, it is dedicated to supplying 
warehouse automation systems to Amazon, which obviously considers them an 
important competitive advantage. 


Most of the examples of automation given above involve manual work. There is one 
occupation which depends almost entirely on cognitive skills which has been largely 
automated out of existence: secretaries. In the 1970s, managers had secretaries, and 
generally did little work on computers themselves. In 1978, secretary was the most 
common job in 21 US states (41% of them). Today, many managers much of their day 
staring at computer screens, and secretary is the most common job in only 4 US states 
(8% of themh [xxiii] 

2.4 - The Luddite fallacy 


Ned Ludd 

A person can have a big impact on society without going to the trouble of actually 
existing. In 1779, Ned Ludd was a weaver in Leicester who responded to being told off 
by his father (or perhaps his employer) by smashing a machine. Or maybe he wasn't - 
the truth is, we don't know. He certainly wasn't the leader of an organised group of 
political protesters. Nevertheless, in the decades following his alleged outburst, his 
name was commonly used to take the blame for an accident or an act of vandalism. 

As Britain pioneered the industrial revolution in the late eighteenth and early 19 th 
century, many of its people attributed their economic misfortune to the introduction of 
labour-saving machines. They were no doubt partly correct, although poor harvests, 
and the Napoleonic Wars against France were also to blame. There was a short-lived 
phenomenon of organised protest under the banner of Luddism in Nottingham in 1811- 
13: death threats signed by King Ludd were sent to machine owners. 

The government responded harshly, with a show trial of 60 men (many of them entirely 
innocent) in York in 1813. Machine breaking was made a capital offence. Riots 
continued sporadically, notably in 1830-31, when the Swing rioters in southern England 
attacked threshing machines and other property. Around 650 of them were jailed, 500 
sent to the penal colony of Australia, and 20 hanged, [xxiv] 

The fallacy 

The Luddites, and other rioters, were not making a general economic or political 
observation that the introduction of labour-saving machinery inevitably causes mass 
unemployment and privation. They were simply protesting against their own dire 
straits, and demanding urgent help from the people who were obviously benefiting. 

It is therefore slightly unfair to them that the term “Luddite fallacy” has become a 
pejorative term for the mistaken belief that technological development necessarily 
causes damaging unemployment. (Although, given the hunger they were experiencing, 
they would probably regard the slur as a very minor irritation.) 

The Luddite fallacy pre-dates the industrial revolution, and has taken in quite a few 

heavyweight thinkers down the years. As long ago as 350BC, the Greek philosopher 
Aristotle observed that if automata (like the ones said to be made by the god 
Hephaestus) became so sophisticated that they could do any work that humans do, then 
workers - including slaves - would become redundant, [xxv] 

During the early 19 th century, when the industrial revolution was in full swing, most 
members of the newly-established social science of economics argued that any 
unemployment caused by the introduction of machinery would be resolved by the growth 
in overall economic demand. But there were prominent figures who took the more 
pessimistic view, that innovation could cause long-term unemployment. They included 
Thomas Malthus, John Stuart Mill, and even the most respected economist of the time, 
David Ricardo. [xxvi] 

The Luddite fallacy and economic theory 

The debate can get quite technical, but there are two reasons why it has been correct to 
reject the Luddite fallacy up until now. The first reason is economic theory: companies 
introduce machines because they increase production and cut costs. This increase in 
supply builds up the wealth in the economy as a whole, and hence the demand for 

Say's Law, named after French economist Jean-Baptiste Say, holds that supply creates 
its own demand, and Say argued that there could not be a “general glut” of any 
particular goods. Of course we do see gluts in sectors of the economy, but an adherent 
of Say's Law would argue these are the unintended consequences of interventions in free 
markets, usually by governments. This law became a major tenet of classical 
economics, but it was rejected emphatically by British economist John Maynard 
Keynes, and is not widely accepted today. 

But many economists would accept a broader interpretation of the law which states that 
reducing the cost of a significant product or service will free up money which was 
previously allocated to it. This money can then be spent to buy more of the item, or 
other items, thereby raising demand generally, and creating jobs. This assumes, 
however, that the money freed up is not spent on expensive assets that generate no 
employment, or invested in companies that employ very few people. 

Economists also point out that the Luddite fallacy also depends on a misapprehension 
about economics called the “Lump of Labour Fallacy”, which is the idea that there is a 
certain, fixed amount of work available, and if machines do some of it then there is 

inevitably less for humans to do. In fact, economies are more organic and more 
flexible: they respond to shifts, and innovate to grow. New jobs are created as old ones 
disappear and the former outnumber the latter. 

The Luddite fallacy and economic experience 

The second reason to reject the Luddite fallacy hitherto is rather better: history has 
proved it to be wrong. A great deal of machinery has been deployed since the start of 
the industrial revolution, and yet there are more people working today than ever before. 
Put simply, if the Luddite fallacy was correct we would all be unemployed by now. 

A study published in August 2015 by the business consultancy Deloitte analysed UK 
census data since 1871 and concluded that far more jobs have been created than 
destroyed by technology in that time. [xxvii] Furthermore, the study argued that the 
quality of the jobs has improved. Where people used to do dangerous and gruelling 
jobs on the land, and hundreds of thousands used to do the work now done by washing 
machines, many more Britons are now employed in caring and service jobs. In the last 
two decades alone there has been a 900% rise in nursing assistants, a 580% increase in 
teaching assistants, and a 500% increase in bar staff- despite the closure of so many of 
the country's pubs. (The authors refrained from commenting on the news that the number 
of accountants has doubled.) 

So in the long run, the Luddite fallacy is just that - a fallacy. But in the short run the 
Luddites had a point. Economists do think that in the first half of the 19 th century, wages 
failed to keep pace with increases in labour productivity. An economist named Arthur 
Bowley observed in the early 20 th century that the share of GDP which goes to labour is 
generally roughly equal to that which goes to capital, [xxviii] but in the first half of the 
19 th century, the share of national income taken by profit increased at the expense of 
both labour and land. The situation changed again in the middle of the century and 
wages resumed their normal growth in line with productivity. It may be that the 
slippage in wages was necessary and inevitable to enable enough capital to be 
accumulated to fuel the investment in technological change. 

The period in the early 19 th century when wage growth lagged productivity growth is 
known as the Engels pause, after the German political philosopher Friedrich Engels, 
who wrote about it in the 1848 “Communist Manifesto”, which he co-authored with 
Karl Marx. The effect ceased at pretty much the same time as he drew attention to it, 
which may explain why it is not better known, [xxix] 

Even in the long run, the picture is not all rosy. A French economist named Gilles 
Saint-Paul has developed a formula which shows that while demand for unskilled 
human labour declines, the demand for skilled human capital increases faster. But a 
side effect can be the increase in income inequality, [xxx] 

Is it different this time? 

Mechanisation and automation has displaced workers on a huge scale since the 
beginning of the industrial revolution. It has imposed considerable suffering on 
individuals, but has led to greater wealth and higher levels of employment overall. The 
question today is whether that will always be true. As machines graduate from offering 
just physical labour to offering cognitive skills as well, will they begin to steal jobs that 
we cannot replace? If the second half of the 19 th century saw “peak horse” in the 
workplace, will the first half of the 20 th century see “peak human”? In other words, is it 
different this time? 

3 - Is it different this time? 

In this chapter I will argue that the arrival of machine intelligence is also the arrival of 
different kind of automation, which spells the end of paid work for many or most 

We will start in section 3.1 by looking at the most popular books to argue the case for 
technological unemployment; we will see how they shy away from the logical 
conclusion of their arguments. We will also hear support for their argument from a 
couple of unexpected sources. In section 3.2 we will briefly review some of the 
academic studies, before hearing in section 3.3 from some sceptics: people who think 
this talk of widespread joblessness is simply the Luddite fallacy at work. 

In later sections of this chapter we will explore in more detail the reasons to believe 
that it really is different this time. 

3.1 - Prophets of change 
Martin Ford 

Martin Ford is the author of perhaps the best book published so far about artificial 
intelligence causing technological unemployment. “The Lights in the Tunnel” (2009) 
provoked fierce debate, and his follow-up, “Rise of the Robots” (2015) fleshed out his 
arguments, and responded to the criticism which the first book attracted. Awarding it 
the 2015 Financial Times and McKinsey Business Book of the Year, Lionel Barber, the 
Financial Times' editor, called it “a tightly-written and deeply-researched addition to 
the public policy debate ... The judges didn’t agree with all of the conclusions, but 
were unanimous on the verdict and the impact of the book.” 

Ford is well-placed to talk about what technology will do to the world of work. He has 
a quarter-century of experience in software design, and he lives and works in Silicon 
Valley, where he runs a software development company. His writing is calm and 
measured, with an engaging humility. 

Exponentials and automation 

Ford opens “Rise of the Robots” with a dramatic illustration of the power of 
exponential increase - the cumulative doubling which is driving digital innovation. He 
asks us to consider driving a car at five miles an hour, and then doubling our speed 27 
times over. The resulting speed would be 671 million miles an hour - fast enough to 
travel to Mars in five minutes [xxxi] . This, he points out, is the number of doublings that 
computer power has gone through since the invention of the integrated circuit in 1958. 
This doubling phenomenon is known as Moore's Law, after one of the founders of the 
chip manufacturer Intel. We will return to this exponential growth later in this chapter, 
as understanding it is fundamental to comprehending the scale of the changes that are 
coming our way. 

The book argues that AI systems are on the verge of wholesale automation of white 
collar jobs - jobs involving cognitive skill such as pattern recognition and the 
acquisition, processing and transmission of information. In fact it argues that the 
process is already under way, and that the US is experiencing a jobless recovery from 
the Great Recession of 2008 thanks to this automation process. Ford claims that middle 
class jobs in the US are being hollowed out, with average incomes going into decline, 

and inequality increasing. He acknowledges that it is hard to disentangle the impact of 
automation from that of globalisation and off-shoring, but he remains convinced that Ai¬ 
led automation is already harming the prospects of the majority of working Americans. 

In fact, since Ford’s book was published the US employment figures have improved 
considerably, and the unemployment rate hovers around 5%, which is considered close 
to full employment. However, many middle-class Americans do feel squeezed, having 
been obliged to accept part-time work, or having missed out on wage rises. This 
suggests that technological unemployment has not yet begun to really bite, but we might 
be seeing the early warning signs. [xxxii] 

Ford pauses to review the prospects for disruption of two sectors of the economy which 
have so far been relatively unscathed by the digital revolution - education and 
healthcare. Although there is fierce resistance to the replacement of human activity by 
AIs in these areas - for instance in essay marking - Ford argues that no industry can 
ignore for long the benefits of cheaper, faster, more reliable ways of providing their 
products and services. He goes on to point out that the companies and industries which 
today are nascent and fast-growing, and tomorrow will be economic giants, are 
extremely parsimonious employers of humans. AirBnB, the peer-to-peer rooms rental 
business, for example, achieved a market cap of $20bn in March 2015 with just 13 

The challenge of UBI 

The final chapters of “Rise of the Robots” explore the consequences of the trends which 
Ford has described. Can an economy thrive and grow if a large minority of people 
cannot find sufficient work to give themselves and their families a decent life? Would 
the consequent rise in inequality be economically harmful? More fundamentally, how 
will these unemployed or under-employed people make ends meet? To Ford, the 
answer to this last question is clear: governments will need to raise the taxes paid by 
those who are still working to provide an income for those who are not. But he is 
acutely aware of the political difficulties that this proposal faces: “American politicians 
are terrified to even utter the word 'tax' unless it is followed immediately by the word 
'cuf .’’ [xxxiii] 

In fact, Ford seems daunted by the situation: “The political environment in the United 
States has become so toxic and divisive that agreement on even the most conventional 
economic policies seems virtually impossible,” he writes. “A guaranteed income is 
likely to be disparaged as 'socialism'”, and “The decades-long struggle to adopt 
universal health coverage in the United States probably offers a pretty good preview of 

the staggering challenge we will face in attempting to bring about any whole-scale 
economic reform.” 

Ford thinks that most people will probably still be able to find some form of paid 
employment - just not enough to make a decent living. Unwilling to give up on 
traditional American ideals like the free market, a capitalist economy and indeed the 
Protestant work ethic, he advocates a universal basic income of only $10,000 a year - a 
level low enough to leave the incentive to find work in place. Even so, he is 
pessimistic about the prospect of persuading his fellow Americans to adopt the idea: “a 
guaranteed income will probably remain unfeasible for the foreseeable future.” 

Andrew McAfee and Erik Brynjolfsson 

As a pair of MIT professors [xxxiv] , McAfee and Brynjolfsson bring academic 
credibility to their book on AI automation, “The Second Machine Age”. They have 
helped to validate the discussion of the possibility of technological unemployment. 

Their book (and their argument) is in three parts. The first part (chapters 1 to 6 
inclusive) describes the characteristics of what they call the second machine age. They 
warn readers that their recitation of recent and forthcoming developments may seem like 
science fiction, and their prose is sometimes slightly breathless: even tenured 
professors can get excited about the speed of technological change and the wonders it 

The Bounty and the Spread 

The second part of the book (chapters 7 to 11) explores the impact of these changes, and 
in particular two phenomena, which they label “bounty” and “spread”. “Bounty” is the 
“increase in volume, variety and quality, and the decrease in cost of the many offerings 
brought on by technological progress. It's the best economic news in the world today.” 
This part of the book could have been written by Peter Diamandis, author of 
“Abundance” and “Bold”, and a leading evangelist for the claim that the exponential 
growth in computer power is leading us towards utopia. 

“Spread” seems to be a synonym for inequality, although the authors are strangely 
reluctant to use that word. jxxxv] It is “ever-bigger differences among people in 
economic success”. This part of the book could have been written by a member of the 
Occupy movemen t[xxxvi] . “Spread is a troubling development for many reasons, and 
one that will accelerate in the second machine age unless we intervene.” 

Brynjolfsson and McAfee pose the question whether bounty will overcome the spread. 

In other words, will we create an economy of radical abundance, in which inequality is 
relatively unimportant because even though a minority is extraordinarily wealthy, 
everyone else is comfortably off? Their answer is that current evidence suggests not. 
Like Martin Ford, they think the American middle class is going backwards financially, 
and they think this trend will continue unless remedial action is taken. 

Hanging onto work 

So the third and final part of the book discusses the interventions which could maximise 
the bounty while minimising the spread. In particular, Brynjolfsson and McAfee want to 
answer a question they are often asked: “I have children in school. How should I be 
helping them prepare for the future?” [xxxvii] They are optimistic, believing that for 
many years to come, humans will be better than machines at generating new ideas, 
thinking outside the box (which they call “large-frame pattern recognition”) and 
complex forms of communication. They believe that humans' superior capabilities in 
these areas will enable most of us to keep earning a living, although they think the 
education system needs to be re-vamped to emphasise those skills, and downplay what 
they see as today's over-emphasis on rote learning. They praise the Montessori School 
approach of “self-directed learning, hands-on engagement with a wide variety of 
materials ... and a largely unstructured school day.” They also have high hopes for 
digital and distance learning, which use “digitisation and analytics to offer a host of 
improvements. ’ ’ [xxxviii] 

Brynjolfsson and McAfee offer a series of further recommendations which they say are 
supported by economists from across the political spectrum: pay teachers more, 
encourage entrepreneurs, enhance recruitment services, invest in scientific research and 
infrastructure improvements, encourage immigration by the world's talented migrants, 
and make the tax system more intelligent. 

These seem somewhat unremarkable proposals, and the authors acknowledge that their 
effectiveness may peter out as the 2020s progress, and machines become even smarter. 
Looking further ahead, they warn against any temptation to try to arrest the progress in 
AI, and also against any temptation to move away from the tried and tested economic 
system of capitalism, which they claim (paraphrasing Churchill's quip about democracy) 
“is the worst form of [economy] except for all the others that have been tried. ” [xxxix] 

The authors are very keen on Voltaire's dictum that “work saves a man from three great 
evils: boredom, vice and need.” They are therefore wary of universal basic income, 

believing that an absence of work will engender boredom and depression. Instead, they 
argue for a negative income tax, which incentivises work. With a negative income tax 
of 50%, if you earn a dollar, the government gives you an additional 50 cents. They cast 
around for ways to keep us all in work, and rather tentatively suggest a range of exotic 
schemes, such as a cultural movement to prefer goods made by humans rather than 

Other voices 

Richard and Daniel Susskind 

Father and son team Richard and Daniel Susskind published “The Future of the 
Professions: How Technology Will Transform the Work of Human Experts” in October 
2015. Richard Susskind has impressive credentials, having worked in legal technology 
since the early 1980s, advised numerous government and industry bodies, and garnered 
a clutch of honorary fellowships If om prestigious universities ,[xl] Perhaps even more 
impressive is that he seems to have retained the respect of his subjects while lambasting 
them as inefficient, and doomed to extinction. 

The Susskinds describe the “grand bargain” whereby members of the professions 
(lawyers, doctors, architects, etc.) enjoy a lucrative monopoly over the provision of 
certain kinds of advice in return for policing the standard of that provision. They argue 
that this bargain has broken down, with many professional services now being available 
only to the rich and well-connected. They demonstrate how important this is by 
illustrating the size of the professions. Healthcare in the US alone now costs $3 trillion 
a year - more than the GDP of the world's fifth-largest country. The combined revenue 
of the Big Four accounting firms, at $ 120 billion, is greater than the GDP of the sixtieth- 
largest country. 

Based on 30 years’ experience of the legal industry and backed up by extensive 
research, they paint two scenarios for its mid-term future. The first has professionals 
working closely with technology, their services enhanced as it improves. The second 
has most or all of the traditional tasks of professionals carried out by machines. The 
Susskinds believe that this second outcome is the inevitable one, since what the rest of 
society cares about is not interaction with humans, but getting our legal, medical and 
other problems sorted out with the minimum of fuss, risk and expense. 

The Susskinds keep their focus on the professions, and refrain from making the obvious 
read-across to the economy as a whole. As a result, they have little to say about 

universal basic income, or the possibility of society fracturing. But they do note that 
once machines have taken on responsibility for most or all the tasks previously carried 
out by human professionals, big questions will be asked about who should own the 
machines. They don't provide answers to these questions, although they indicate their 
preference for some form of common ownership which does not involve the state. In 
this respect they deserve credit for following the logic of their arguments further than 
most people writing on the subject. 

The book is written with refreshing clarity, precision and felicity of expression - and 
with such a gloomy message for its audience, that is probably just as well. 

Scott Santens 

Scott Santens is a writer and a campaigner for Universal Basic Income, based in New 
Orleans .[xli] He is a moderator of the Reddit Basic Income page, where he maintains a 
useful FAQ on the subject. [xlii] Self-employed since 1997, towards the end of 2015 he 
managed to procure a basic income for himself based on pledges from others who 
support his campaign, via the online giving site Patreon. 

Jerry Kaplan 

Serial entrepreneur Jerry Kaplan co-founded GO Corporation, which was a precursor 
to smartphones and tablets, and was sold to AT&T. He also co-founded OnSale, an 
internet auction site which pre-dated Ebay, and was sold for $400m. He teaches at his 
alma mater, Stanford University, and writes books, including one called “Humans Need 
Not Apply”. 

Its message is similar to “The Second Machine Age”: AI has reached a tipping point 
and is becoming powerfully effective. This will disrupt most walks of life (the 
computer, he observes, is blind to the colour of your collar), and unless we manage the 
transition well, the resulting economic instability and growing inequality could be 

Like Ford, Brynjolfsson and McAfee, Kaplan thinks the existing market economy can 
survive this transition intact. 

CGP Grey 

Kaplan got the title “Humans Need Not Apply” from a video of the same name [xliii] 
which appeared on the internet a year before. Posted to YouTube in August 2014 by an 
Irish-American who goes by the name CGP Grey (his full name is Colin Gregory 
Palmer Gre v[xliv] ). the video attracted over 5 million views within a year. 

The video is well-produced, engaging and persuasive. It contains plenty of 
technological eye-candy, and makes its points in punchy sound-bites - ideal for today's 
short attention spans. Unlike the books described above, it offers no solutions to the 
problems raised by AI and robotic automation, but - also unlike them - it suggests that 
capitalism cannot cope with what is coming. 

Gary Marcus 

A psychology professor at New York University, Gary Marcus has taken an intense 
interest in artificial intelligence, and where it is leading us. In February 2015 he told a 
CBS interviewer “Eventually I think most jobs will be replaced, like 75-80% of people 
are probably not going to work for a living... There are a few people starting to talk 
about it.” [xlv] 

Federico Pistono 

Federico Pistono is a young Italian lecturer and social entrepreneur. He attracted 
considerable attention with his 2012 book “Robots Will Steal Your Job, But That's 
OK”. A range of eminent people, including Google's Larry Page, were drawn to its 
optimistic and discursive style. (Google re-named itself Alphabet in October 2015, but 
most people still call it Google, so in this book I’ll mostly follow that convention.) 

After making a forceful case that future automation will render most people 
unemployed, Pistono argues that there is no need to worry. Much of the book is taken up 
with musing on the nature of happiness - the word features in the titles of a quarter of its 
chapters. He is hopeful that we will all discover that the pursuit of happiness through 
material goods is a fool's errand, and he argues that salvation lies in downsizing. He 
offers the example of his own family, living in northern Italy. They spend $45,000 a 
year, but by getting rid of two of their three cars, growing their own food, and 
generating their own electricity, they can reduce this to $29,000 a year. 

He also urges us all to educate ourselves - and encourage everyone else to do likewise 
- but more for personal fulfilment than in a vain attempt to remain employable. 

Two unexpected voices 
Andy Haldane 

As the chief economist of the Bank of England, Andy Haldane isn't the most obvious 
person to be found musing about the benefits of universal basic income. But that is 
exactly what he did in a speech at the Trades Union Congress in November 2015 [xlvi] . 
He wondered whether the displacement effect of automation, whereby jobs are 
destroyed, might start to outweigh the compensation effect, whereby automation raises 
productivity sufficiently to generate more demand and thus work. 

In his speech, Haldane avoided giving a definitive answer to the question of whether we 
are nearing “peak human”, but he raised many of the concerns explored in this book. He 
presented an estimate prepared by the Bank of England of the likelihood of automation 
of the jobs in a range of economic sectors in the UK, adapted from the estimates 
produced for the US by Frey and Osborne of the Oxford Martin School (of which, more 
below). The Bank estimated the UK's situation as slightly less alarming than that of the 
US, but not much. It found that roughly a third of jobs have a low probability of being 
automated out of existence, another third have a medium probability, and the final third 
have a high probability. Haldane avoided putting a specific timescale on this, and also 
avoided saying what would happen after that undisclosed period. 

Martin Wolf 

As the main financial columnist and associate editor at the Financial Times, Martin 
Wolf is the very epitome of a City establishment figure. He was described by US 
Treasury Secretary Larry Summers as “probably the most deeply thoughtful and 
professionally informed economic journalist in the world.” [xlvii] Although the credit 
crunch and subsequent recession have re-kindled his youthful enthusiasm for Keynesian 
economics, it is still a surprise to read him advocating income redistribution and 
universal basic income, as he did in this article from February 2014: 

“If Mr Frey and Prof Osborne [see below] are right [about automation]... we will need 
to redistribute income and wealth. Such redistribution could take the form of a basic 
income for every adult, together with funding of education and training at any stage in a 
person’s life. ... The revenue could come from taxes on bads (pollution, for example) or 
on rents (including land and, above all, intellectual property). Property rights are a 
social creation. The idea that a small minority should overwhelmingly] benefit from 
new technologies should be reconsidered. It would be possible, for example, for the 

state to obtain an automatic share in the income from the intellectual property it 
protects .’’ [xlviii] 

3.2 - Academic and consultancy studies 

Numerous reports have been written about technological unemployment by academic 
organisations, consultancies, and think tanks. I have described some of the better- 
known ones here. Sometimes they reserve judgement or sit on the fence, but as far as 
possible, I present them in order of increasing scepticism about the proposition of 
widespread unemployability. 

Frey and Osborne 

Carl Benedikt Frey and Michael Osborne are the directors of the Oxford Martin 
Programme on Technology and Employment, [xlix] Their 2013 report “The future of 
employment: how susceptible are jobs to computerisation?” has been widely quoted. 

Its approach to analysing US job data has since been used by others to analyse job data 
from Europe and Japan. 

The report analyses 2010 US Department of Labour data for 702 jobs, and in a curious 
blend of precision and vagueness, concludes that “47% of total US employment is in the 
high risk category, meaning that associated occupations are potentially automatable over 
some unspecified number of years, perhaps a decade or two.” 19% of the jobs were 
found to be at medium risk and 33% at low risk. Studies which have extended these 
findings to other territories have yielded broadly similar results. 

The methodology overlays rigour on guesswork. 70 of the jobs were categorised in a 
brainstorming session, and these categorisations were then extended to the other 632 
jobs using calculations which will mystify anyone with only school-level maths, 
including Gaussian process classifiers - a statistical tool also used in deep learning AI 
systems. But it would be unfair to criticise the report for lack of rigour. Forecasting is 
not an exact science; the authors adopted the most scientific approach they could devise, 
and made no attempt to hide its subjective elements. 

As well as sounding the alarm about the possibility of technological unemployment, the 
report suggests that the “hollowing out” of middle class jobs will stop. A 2003 paper 
by David Autor (of whom, more below) observed that income has increased for high 
earners and (albeit less rapidly) for low earners, but stagnated for medium-level 
earners. Maarten Goos and Alan Manning characterised this hollowing out as the 
favouring of “Lovely and Lousy jobs”. Frey and Osborne argue that in the future, 
susceptibility to automation will correlate negatively with income and educational 

attainment, so the Lousy jobs will also disappear. They suggest that people will have to 
acquire creative and social skills to stay in work, but they don't appear to think that 
many of us will be able to change the fate that our employment history has assigned to 

Frey and Osborne followed the 2013 report with another in February 2015, written in 
collaboration with senior bankers from Citibank. It provides insights into the impact of 
automation in a number of industry sectors, including stock markets, where the move 
from trading floors to digital exchanges reduced headcount by 50%. At first glance it is 
surprising to see bankers suggesting increased taxation to provide income for the 
unemployed, but it also seems they have little faith in it happening: “Such changes in 
taxation would seem sensible to us, but they would also be a reversal of the trends of 
the last few decades.” They don't hold out much more hope for their other principal 
suggested remedy: “education alone is unlikely to solve the problem of surging 
inequality, [but] it remains the most important factor.” 


Gartner is the world’s leading technology market research and advisory consultancy. At 
its annual conference in October 2014, its research director Peter Sondergaard declared 
that one in three human jobs would be automated by 2025. [1] "New digital businesses 
require less labor; machines will make sense of data faster than humans can." He 
described smart machines as an example of a “super class” of technologies which carry 
out a wide variety of tasks, both physical and intellectual. He illustrated the case by 
pointing out that machines have been grading multiple choice examinations for years, but 
they are now moving on to essays and unstructured text. 

The Millennium Project 

The Millennium Project was established in 1996 by a coalition of UN organisations and 
US academic research bodies. Its “2015-16 State of the Future” contained a section on 
the future of work based on a poll of 300 experts from around the world. Although they 
mostly thought that technology would impact employment significantly, their collective 
estimates for long-term unemployment were relatively conservative. They expected 
global unemployment to reach only 16% in 2030, and just 24% in 2050. 

Pew Research Center 

The Pew Research Center published a report entitled “AI, robotics, and the future of 
jobs”[li] in November 2014. The Center is part of the Pew Charitable Trusts, 
established in 1948 with over $5bn bequeathed by descendants of the founder of Sun 
Oil; the Center is the third-largest think tank in the US. 

The Center sent a questionnaire to 12,000 selected experts and interested members of 
the public (mostly but not entirely American), and received 1,900 responses to the 
question “Will networked, automated, artificial intelligence (AI) applications and 
robotic devices have displaced more jobs than they have created by 2025?”. A slight 
majority (52%) said no, arguing that technology has always created more jobs than it 
has destroyed, that it is not advancing fast enough to destroy so many jobs, and that 
regulatory intervention would stop it if necessary. 

The 48% who thought there would be a net loss of jobs believed that the process was 
already in train, but that it would get much worse, and that inequality would become a 
severe problem as a result. 

Both sides thought that the education system is doing a poor job of preparing young 
people for the new world of work, and also that the future of employment is not pre¬ 
ordained, but is susceptible to good policy. 

Fundacion Innovacion Banklnter 

Banklnter, based in Madrid, is one of the largest banks in Spain. In 2003 it established 
a Foundation to promote the creation of sustainable wealth in Spain through innovation 
and entrepreneurship. One of the Foundation's main activities is organising the Future 
Trends Forum, an international think tank which periodically gathers together a group of 
experts to discuss an important topic, and then produces reports and videos based on the 
conclusions of those discussions. 

In June 2015 I took part in a meeting of the Future Trends Forum entitled “The Machine 
Revolution”, which addressed “how technological developments (internet, robotics, 
artificial intelligence, etc.) will boost employment and labour markets in the next 
decade.” Compered by Chris Meyer, author of “Standing on the Sun”, the delegates 
were a mixture of senior government officials from around the world, academic and 
commercial economists, investors and writers. 

When the 34 experts at the meeting were asked whether we thought structural 
unemployment was a likely result, a slight majority said not. Towards the end of the 

meeting we each contributed two predictions to a collective timeline, which appears at 
the end of the report, [lii] 


The world's most prestigious management consultancy firm weighed in on the subject of 
technological unemployment with an article published in its quarterly magazine in 
November 2015 entitled “Four fundamentals of workplace automation” [liii] . Billed as 
an interim report of an ongoing research project, its central argument was that instead of 
asking which jobs can be and will be automated, we should ask which tasks will be 
automated. Few people, it claimed, will find that their entire job disappears, but as 
much as 45% of the tasks people do at work can be automated with technology that is 
currently available. 

The McKinsey consultants identified 2,000 different “activities” (e.g., greeting 
customers, demonstrating product features) for a selection of “occupations” (e.g., retail 
salesperson), and assessed which activities required the 18 “capabilities” which they 
deem susceptible to automation (e.g., understanding natural language, generating natural 
language, retrieving information). 

They noted that the level of automatability will rise as machines become more capable. 
For instance, if and when machines equal the median human level of natural language 
comprehension, then the proportion of tasks which can be automated will rise from 45% 
to 58%. 

At the time of publication, the authors concluded that only 5% of jobs were capable of 
being fully automated, but 60% of jobs could have 30% of their constituent activities 
automated. But rather than leading to a 30% headcount reduction, with the other 70% of 
activities being smeared among the remaining employees, they expect people to become 
more productive, as machines augment human performance. 

Very highly-paid jobs tended to have fewer automatable activities (e.g., 20% for 
CEOs), but among low- and medium-paid jobs there was a fairly even spread. The 
consultants found that only 4% of the work activities carried out in the US require 
creativity at the median human level, and only 29% require a median human level of 
emotion sensing. Optimistically, they concluded from this that automation will enable 
humans to do better and more interesting work. Interior designers, for instance “could 
spend less time taking measurements, developing illustrations and ordering materials, 
and more time developing innovative design concepts.” 

Finally, McKinsey suggested that senior managers should increasingly pay close 
attention to the type, direction and potential of automation within their industry, as it 
will become a more and more important source of competitive advantage. 

A swelling chorus 

The reports described above are a selection of the most prominent ones published so far 
on the subject of technological unemployment. They are not the only ones, and more are 
being produced every month - sometimes every week. There is no clear consensus 
about the likely impact on joblessness of machine intelligence in the coming years and 
decades. Nevertheless, the theme has an increasingly high profile in the media - it was 
a focus of the annual gathering of the super-rich and powerful in the ski resort of Davos 
in January 2016. 

In the next section we will see that some people are firmly convinced it is a myth. 

3.3 - Crying wolf 

In this section we meet a selection of writers who are sceptical about the prospect of 
technological unemployment. They argue that it is all just a revival of the Luddite 

David Autor 

David Autor is a professor of economics at MIT. As noted above, he sounded the alarm 
in a 2003 paper about the “hollowing out” of middle class jobs in the USA - the fact 
that income has increased for high earners and (albeit less rapidly) for low earners, but 
stagnated for medium-level earners. 

In an interview in October 2015 [liv] . he gave three reasons why he thinks that some 
observers have been unduly pessimistic, even hysterical, about the likelihood of job 
destruction. One is that machines complement and augment humans: they always have, 
and there is no reason to think that is about to change. The second is that machines 
increase productivity, which creates wealth, consumption; and demand, which creates 
more jobs. 

The third reason is that humans are creative and ingenious. There are many important 
businesses and activities now that could not have been imagined 50 years ago. In fact, 
Autor accuses Martin Ford of arrogance in writing off human ingenuity. 

In an article for the Journal of Economic Perspectives (summer 2015) entitled “Why are 
there still so manvjobs?”.[lv] Autor forecasts that people will retain a comparative 
advantage in so-called “human” attributes such as interpersonal interaction, flexibility, 
adaptivity. He argues that many jobs - like radiologist - combine these with the 
routine, predictable tasks where computers win. Autor believes it will not be possible 
to separate these two types of tasks, so humans will continue to carry out the whole 

More generally, Autor is also one of those who believe the rate of change today is over¬ 
hyped. He believes that the effect of Moore's Law is substantially muted by regulatory 
and social frictions which slow down the adoption of new technologies, and he also 
argues that many technological advances simply don't translate into tangible 
improvements in the real world. For instance he accepts that his current computer is a 
thousand times faster than the one he used a few years ago, but he suspects it only makes 

him 20% more productive. It may be true, he teases, that a new washing machine has 
more processing power than NASA used to send Neil Armstrong to the moon in 1969, 
but the washing machine is still not going to the moon. 

He derides some of the heralded achievements of AI researchers, arguing for instance 
that self-driving cars do not emulate human drivers, but instead rely on precise maps of 
the terrain which have to be prepared before the journey starts. This makes them less 
flexible than humans, and not fit to be released into the wild without human escorts. We 
will see how well these arguments stand up later in this chapter. 

Although Autor is broadly optimistic about our future, he believes that much depends on 
the decisions that we take. “If machines were in fact to make human labour superfluous, 
we would have vast aggregate wealth but a serious challenge in determining who owns 
it and how to share it.” He points out that Norway and Saudi Arabia both enjoy 
economic abundance (thanks to oil rather than AI), but they use it very differently. 
Norwegians, he says, work few hours per day and are generally happy; Saudis inport 
90% of their labour and nurture terror. 

Robin Hanson 

Robin Hanson is an associate professor of economics at George Mason University, in 
Virginia, USA. Like David Autor, Hanson castigates Martin Ford for inappropriate 
motives, but whereas Autor accuses Ford of arrogance, Hanson alleges dishonesty: “In 
the end, it seems that Martin Ford's main issue really is that he dislikes the increase in 
inequality and wants more taxes to fund a basic income guarantee. All that stuff about 
robots is a distraction. ”JTvi] 

After a few more jibes, Hanson addresses Ford's actual thesis. He starts by admitting 
that “Ford is correct that, ... in the long run, robots will eventually get good enough to 
take pretty much all jobs. But why should we think something like that is about to 
happen, big and fast, now?” He attributes four arguments to Ford, and makes short work 
of the first three. The first is the Frey and Osborne study we reviewed in chapter 3.2, 
which Hanson dismisses as subjective. The second argument is the decline in labour's 
share of income since 2000, which Hanson replies could be caused by numerous other 
factors rather than technological automation. The third argument is the rapid fall in 
computer prices, which Hanson says has yet to cause any detectable unemployment. 

“And then there is Ford's fourth reason: all the impressive computing demos he has seen 
lately.” Hanson is referring, of course, to Google's self-driving cars, real-time machine 
translation systems, IBM's Watson and so on. Hanson is less impressed by these 

demonstrations of rapidly improving AI: “We do expect automation to take most jobs 
eventually, so we should work to better track the situation. But for now, Ford's reading 
of the omens seems to me little better than fortune telling with entrails or tarot cards.” 

Having unburdened himself of this cynicism, Hanson proceeds to offer a constructive 
suggestion. He advocates forecasting by means of prediction markets, where people 
place bets on particular economic or policy outcomes, like the level of unemployment at 
some future date. He argues that prediction markets give us a financial stake in being 
accurate when we make forecasts, rather than just trying to look good to our peers. 

Tyler Cowen 

A professor at George Mason University and co-author of an extremely popular blog, 
Tyler Cowen was New Jersey's youngest ever chess champion. He is a man with 
prodigiously broad knowledge and interests, and although he proposes some key ideas 
forcefully, there is always some nuance, and he dislikes simplistic and modish 
solutions. In two recent books, “the Great Stagnation” (2011) and “Average is Over” 
(2014), he paints a picture of America's future which is slightly depressing, but not 
apocalyptic. He is alive to the prospect of dramatically improved AI, and the effect it 
will have on employment. But he does not think widespread permanent unemployment 
will be one of its results. 

For several years, Cowen has championed the claim that the US economy is hollowing 
out. He expects automation to continue this trend, perhaps to accelerate it. In an article 
for Politico magazine [lvii] . he wrote: 

“I imagine a world in which, say, 10 to 15 percent of the citizenry ... is extremely 
wealthy and has fantastically comfortable and stimulating lives, equivalent to those of 
current-day millionaires, albeit with better health care. Much of the rest of the country 
will have stagnant or maybe even falling wages in dollar terms.” This grim outlook for 
the majority is softened because “they will have a lot more opportunities for cheap fun 
and cheap education [thanks to] all the free or nearly free services that modern 
technology makes available.” But there is a sting in the tail for the real underclass. 

They, he says, “will fall by the wayside.” 

Cowen does not expect a universal basic income to be required. Nor does he expect 
riots. One reason is that the US population will be older: “By 2030, about 19 percent of 
the US population will be over 65; in other words, we’ll be as old as Floridians are 
today.” Floridians are a conservative lot, not given to mayhem. Another is that people 
will increasingly cluster geographically according to income. Few people in the poorer 

85% will live in the hothouse cities of San Francisco and New York, and they will not 
have the wealth of Manhattan waved in their faces. And perhaps most important, the 
masses will inure themselves with the opiates of tree entertainment and social media. 

Geoff Colvin 

Geoff Colvin is an editor at Fortune magazine and one of America's most experienced 
and respected journalists. In August 2015 he published “Humans Are Underrated: What 
High Achievers Know That Brilliant Machines Never Will.” His previous book, 
“Talent is over-rated” (2006) advanced the proposition that hours of dedicated practice 
trumps talent in most endeavours, and it was an international best-seller. 

His new book accepts that for the first time, technology may be reducing total 
employment rather than increasing it, but is sceptical for two reasons in particular. 

First, Colvin argues that because it is so hard to foresee the new types of jobs that are 
created when economies shift (just as web development and social media marketing 
were hard to foresee), we under-estimate how many of them there will be. 

Colvin believes that skills of deep human interaction - empathy, storytelling, the ability 
to build relationships - will become far more valuable in the future, and many people 
will be able to prosper by bringing those skills into the evolving economy. “We’re 
hard-wired by 100,000 years of evolution to value deep interaction with other humans 
(and not with computers). Those wants won’t be changing anytime soon.” [lviii] 

Crying wolf 

“The boy who cried wolf’ is one of Aesop’s fables, and a commonly-told children’s 
story. The moral usually drawn is that people who earn a reputation for lying are later 
punished by being disbelieved, [lix] But the story has another lesson for us: a claim that 
was false in the past may be true in the future, and it can be dangerous to forget that. 
Automation has been going on for centuries, and past claims that it was causing 
permanent widespread unemployment have been proven wrong. So far. We should not 
be complacent when there are good reasons to think that this time, it may be different. 

3.4 - AI to date 

We now need to spend some time looking at the advances in artificial intelligence 
which have prompted the discussion we have just witnessed. 

If you have read my previous book, “Surviving AI”, parts of the next two sections may 
seem familiar. (And by the way, bless you.) They do contain updates, as the field is 
moving fast. We get back to new territory in chapter 3.6. 

What is AI? 

Intelligence is the measure of an agent’s ability to achieve goals in a range of different 
environments. [lx] In both humans and machines, intelligence is not a single, unitary 
phenomenon. American psychologist Howard Gardner has distinguished nine types of 
human intelligence: linguistic, logic-mathematical, musical, spatial, bodily, 
interpersonal, intra-personal, existential and naturalistic. As you read that list, you are 
probably thinking that you are better at some than others, and that you know other people 
whose mixture of skills is different. It is the same with artificial intelligence. 

Artificial intelligence is simply an intelligence that did not arise naturally by evolution, 
but was created by humans (or perhaps aliens). Many people think the term is 
unfortunate and temporary: cars are not called artificial horses, and planes are not 
called artificial birds. They prefer terms like machine intelligence, or cognitive 
computing. I sympathise, although for the moment at least, the term artificial 
intelligence, or AI, is the one understood by the broadest range of people. I will use the 
terms machine intelligence and artificial intelligence as synonyms. 

The value of intelligence 

Intelligence is, of course, the distinguishing feature of humans: it is the characteristic 
which sets us apart from other animals and makes us more powerful than them. And we 
are much, much more powerful than them Genetically, we are almost identical to 
chimpanzees, and our brains are not much heavier than theirs per kilo of body weight. 
But the difference in structure between our brains means that there are 7 billion of us 
and only a few hundred thousand of the m[lxi] . Their fate depends on our actions, and 
they are not even aware of that fact. 

Our intelligence enables us to communicate, to share information and ideas, and to 
devise and execute plans of action. It also enables us to develop tools and technology. 

A single unarmed human would be slaughtered by a mammoth or a lion, but a group of 
humans working together, or a single human equipped with a rifle, can turn the tables 
very effectively. 

Before we go any further we need to distinguish between intelligence and 
consciousness, and note that the former does not seem to require the latter. Ins ects 
display a level of intelligence - especially collectively - but we have no evidence that 
they are conscious to any significant extent. Intelligence and consciousness are both 
more in evidence among mammals, especially primates, but there does not appear to be 
a straight-line correlation between the two. 

Machine intelligence is unlike animal intelligence in that machines can be super-human 
in very narrow fields - like performing mathematical calculations, or playing chess - 
but utterly unintelligent in all other respects, and (so far as we can tell) completely 
lacking in any degree of consciousness. 

We value intelligence highly, since it is the source of our power, but we value 
consciousness even more. Most humans are happy to kill and eat animals which we 
deem to have a lower level of consciousness than our own. There is no reason to 
suppose that humans have attained anywhere near the maximum possible level of 
intelligence, and it seems highly probable that we will eventually create machines that 
are more intelligent than us in all respects - assuming we don't blow ourselves up first. 
We don't yet know whether those machines will be conscious, let alone whether they 
will be more conscious than us - if that is even a meaningful question. 

Artificial General Intelligence (AGI) and Superintelligence 

As we noted in chapter 1, the term for a machine which equals or exceeds human 
intelligence in all respects is artificial general intelligence, or AGI. The day when the 
first such machine is built will be a momentous one, as the arrival of superintelligence 
will not be far beyond it. The likelihood of an intelligence explosion is commonly 
referred to as the technological singularity. This could be an astonishingly positive 
development for humankind, or a disastrously negative one. 

I wrote about this extensively in my previous book, “Surviving AI”, and will not cover 
that ground again here. Suffice to say, we should make strenuous efforts to ensure that if 
and when we do create the first machines which are destined become 
superintelligences, we experience a positive outcome rather than a negative one. 

Anders Sandberg of Oxford University’s Future of Humanity Institute summarised it 
well by saying that we should aim to become the mitochondria of superintelligence 
rather than its boot loader. He was referring to Elon Musk’s metaphor for how, if we 
are unwise and / or unfortunate, we could create the thing which destroys us, and saying 
that we should aim instead for the fate of the prokaryotic cell which was absorbed by 
another, larger cell and became an essential component of a new, combined, and more 
complex entity, the first eukaryotic cell. 

This book is concerned with the impact of “narrow” AI systems which fall considerably 
short of AGI. 

A quiet revolution 
Origins and winters 

The science of artificial intelligence got started in 1956 at a conference held at 
Dartmouth College, in New Hampshire. Since then it has gone through cycles of 
optimism and pessimism. Herbert Simon said in 1965 that “machines will be capable, 
within twenty years, of doing any work a man can do.” [lxii] and two years later Marvin 
Minksy said that “Within a generation ... the problem of creating 'artificial intelligence 1 
will substantially be solved.” [lxiii] These early claims turned out to be ill-founded, 
and later generations of researchers found their sources of funding dried up in so-called 
AI winters. 

Some leading figures in the field today are worried that a similar fate may befall them 
because, they say, excessive claims are being made about the capabilities of AI systems 
today, and what can be achieved in the short term. This seems to me an ungrounded 
fear. Machine intelligence is the target of enormous investments - by technology giants 
like Google and Facebook, by startups, by traditional companies like the automotive 
manufacturers, and by governments. These investments are being made because 
machine intelligence delivers results, and they will continue so long as that remains the 
case. It will only stop if the results stop coming. 

Machine learning 

In the last few years, the field of AI has undergone a quiet revolution. It goes by the 
name of machine learning, and a subset called deep learning has proved especially 
effective at tasks which were previously considered hard problems that were unlikely to 

be solved for many years to come. 

The approach to AI which prevailed in its early days tried to reduce human thought to 
the manipulation of symbols, such as language and maths, which could be made 
comprehensible to computers. This became known as symbolic AI, or Good Old- 
Fashioned AI (GOFAI). Machine learning, by contrast, is the process of creating and 
refining algorithms which can produce conclusions based on data without being 
explicitly programmed to do so. The turning point came in 2012 when researchers in 
Toronto led by Geoff Hinton won an AI image recognition competition called 
ImageNet. [lxiv] Hinton is a British researcher now at Toronto University and Google, 
and perhaps the most important figure behind the rise of deep learning as the most 
powerful of today's AI techniques. 

( The word algorithm comes from the name of a 9 th -century Persian mathematician, Al- 
Khwarizmi. [lxv] It means a set of rules or instructions for a person or a computer to 
follow. It is different from a programme, which gives a computer precise, step-by-step 
instructions how to handle a very specific situation such as opening a spreadsheet, or 
calculating the sum of a column of figures. An algorithm can be applied to a wide range 
of data inputs. A machine learning algorithm uses an initial data set to build an internal 
model which it uses to make predictions; it tests these predictions against additional 
data and uses the results to refine the model. The way that some game-playing AIs 
become superhuman in their field is by playing millions of games against versions of 
themselves and learning from the outcomes.) 

In deep learning, the algorithms operate in several layers, each layer processing data 
from previous ones and passing the output up to the next layer. The output is not 
necessarily binary, just on or off: it can be weighted. The number of layers can vary 
too, with anything above ten layers seen as very deep learning - although in December 
2015 a Microsoft team won the ImageNet competition with a system which employed a 
massive 152 layers. [lxvi] 

Deep learning, and especially artificial neural nets (ANNs), are in many ways a return 
to an older approach to AI which was explored in the 1960s but abandoned because it 
proved ineffective. While Good Old-Fashioned AI held sway in most labs, a small 
group of pioneers known as the Toronto mafia kept faith with the neural network 
approach. They were vindicated when it was discovered that applying them to very 
large data sets made them surprisingly effective. 

How good is today's AI? 

Games and quizzes 

The game of chess used to be thought of as one of the most challenging intellectual 
pursuits a person could undertake. (Being rubbish at it, I still do.) It used to be thought 
that it would take centuries for machines to become really good at it. That was a long 
time ago, of course, and we are much wiser now, because as long ago as 1997, IBM's 
Deep Blue beat Gary Kasparov, the world's best player, in a controversial but 
conclusive match. Nowadays, humans have no chance against even mid-level chess 

With the benefit of hindsight, we know we should have expected this. The rebarbative 
MIT professor Noam Chomsky observed that a computer winning a chess competition is 
no more surprising than a forklift truck winning a weightlifting contest. Maybe he was 
saying this long before Deep Blue beat Kasparov, but if so he was unusual. In truth it 
has often been hard to forecast what will be hard for machines to do. It turned out to be 
relatively easy to programme computers to do things that we find very hard, but very 
hard to teach them how to do things that we find easy, like tying our shoelaces. This is 
known as Moravec’s paradox, after AI pioneer Hans Moravec. [lxvii] 

This episode is also a good illustration of another phenomenon, which is that once a 
computer is able to perform a particular task better than humans, we dismiss it as 
simple, saying that the next challenge is the really hard one. Until it isn't. 

In fact, once a machine is able to perform a particular task, we usually stop calling it 
artificial intelligence. This is known as Tesler's Theorem, which defines artificial 
intelligence as that which a machine cannot yet do. 

IBM's next bravura AI performance came in 2011, when a system called Watson beat 
the best human players of the TV quiz game “Jeopardy”, in which contestants are given 
an answer and have to deduce the question. Watson used “more than 100 different 
techniques ... to analyze natural language, identify sources, find and generate 
hypotheses, find and score evidence, and merge and rank hypotheses.” It had access to 
200 million pages of information, including the full text of Wikipedia, but it was not 
online during the contest. The difficulty of the challenge is illustrated by the answer, "A 
long, tiresome speech delivered by a frothy pie topping" to which the target question 
(which Watson got right) was "What is a meringue harangue?" After the game, the 
losing human contestant Ken Jennings famously quipped, “I for one welcome our new 
robot overlords. ” [lxviii] 

At the beginning of this chapter we noted that intelligence is not a single, unitary skill or 

process. The fact that Watson is an amalgam - some would say a kludge - of numerous 
different techniques does not in itself mark it out as different and perpetually inferior to 
human intelligence. It is nowhere near an artificial general intelligence which is human- 
level or beyond in all respects. It is not conscious. It does not even know that it won 
the Jeopardy match. But it may prove to be an early step in the direction of artificial 
general intelligence. 

In January 2016, an AI system called AlphaGo developed by Google's DeepMind beat 
Fan Hui, the European champion of Go, a board game. This was hailed as a major step 
forward: the game of chess has more possible moves (35 80 ) than there are atoms in the 
visible universe, but Go has even more - 250 15 °. [lxix] The system uses a hybrid of AI 
techniques: it was partly programmed by its creators, but it also taught itself using a 
machine learning approach called deep reinforcement learning. 

(Reinforcement learning is halfway between two other important forms of machine 
learning: supervised and unsupervised learning. In supervised learning the system is 
given an example to follow at each step. In unsupervised learning there are no 
examples. In reinforcement learning there are rewards for successful steps and 
penalties for unsuccessful steps. The system has to figure out how to behave according 
to those signals. [lxxj f 

A match against the world champion Lee Se-Dol followed in March 2016. Se-Dol was 
confident, believing it would take a few more years before a computer could beat him. 
He was genuinely shocked to lose the series four games to one, and observers were 
impressed by AlphaGo’s sometimes unorthodox style of play. AlphaGo’s achievement 
was another landmark in computer science, and perhaps equally a landmark in human 
understanding that something important is happening, especially in the Far East, where 
the game of Go is far more popular than it is in the West. 

DeepMind did not rest on its laurels. A month after its European Go victory it 
presented a system able to navigate a maze in a video game without access to any maps, 
or to the code of the game. Using a technique called asynchronous reinforcement 
learning, the system looked at the screen and ran scenarios through multiple versions of 
itself, [lxxi] The ability to navigate by sight, like humans do, will be invaluable for AIs 
in many real-world applications. 

Self-driving vehicles 

Another landmark demonstration of the power of AI began inauspiciously in 2004. 

DARPA offered a prize of $ 1 million to any group which could build a car capable of 
driving itself around 150-mile course in the Mojave Desert in California. The best 
contestant was a converted humvee named Sandstorm which got stuck on a rock after 
only 7miles. [lxxii] Eight years later, Google's self-driving cars have driven well over a 
million miles without being responsible for a single serious accident. It is true that they 
have been rear-ended by human drivers a few times, but this is because they obey traffic 
regulations, and we humans are not used to drivers doing that. A Google car drove into 
a bus on Valentine’s Day 2016, but the facts of that incident remain somewhat 
ambiguous . [lxxiii] 

The world's automotive manufacturers are now scrambling to master the technologies 
involved in producing self-driving cars. Toyota is investing billions of dollars in 
research facilities in Silicon Valley. [lxxiv] and in December 2015, Ford announced a 
joint venture with Google, [lxxv] Elon Musk, CEO of the startlingly impressive upstart 
car manufacturer Tesla, said in December 2015 that its first fully autonomous cars 
would be sold in two years, [lxxvi] Of course it will take longer than that for regulators 
to catch up, and it will take years for enough of the old, human-piloted cars to be 
replaced for self-driving cars to deliver the tremendous benefits they are capable of. 

We will explore this in more detail in chapter 3.8. 


We are strangely nostalgic about the future, and we are often disappointed that the 
present is not more like the future that was foretold when we were younger. 2015 was 
the 30 th anniversary of the 1985 movie “Back to the Future”, and it was also the year to 
which the hero travels at the end of the story. Journalists and commentators complained 
about the failure of hoverboards and flying cars to arrive, as predicted in the film. 

We didn't get hoverboards, but we did get something even more significant. As recently 
as the late 20 th century, knowledge workers could spend hours each day looking for 
information. Today, less than twenty years after Google was incorporated in 1998, we 
have something close to omniscience. At the press of a button or two, you can access 
pretty much any knowledge that humans have ever recorded. To our great-grandparents, 
this would surely have been more astonishing than flying cars. 

(Some people are so impressed by Google Search that they have established a Church 
of Google, and offer nine proofs that Google is God, including its omnipresence, near- 
onmiscience, potential immortality, and responses to prayer, [lxxvii] Admittedly, at the 
time of writing, there are only 427 registered devotees, or “readers”, at their meeting- 

place, a page on the internet community site Rcddit. [lxxviii] i 

In the early days, Google Search was achieved by indexing large amounts of the web 
with software agents called crawlers, or spiders. The pages were indexed by an 
algorithm called PageRank, which scored each web page according to how many other 
web pages linked to it. This algorithm, while ingenious, was not itself an example of 
artificial intelligence. Over time, Google Search has become unquestionably AI- 

In August 2013, Google executed a major update of its search function by introducing 
Hummingbird, which enables the service to respond appropriately to questions phrased 
in natural language, such as, “what's the quickest route to Australia?” [lxxix] It 
combines AI techniques of natural language processing with colossal information 
resources (including Google's own Knowledge Graph, and of course Wikipedia) to 
analyse the context of the search query and make the response more relevant. PageRank 
wasn't dropped, but instead became just one of the 200 or so techniques that are now 
deployed to provide answers. Like IBM Watson, this is an example of how AI systems 
are often agglomerations of numerous approaches. 

In October 2015, Google confirmed that it had added a new technique called RankBrain 
to its search offering. RankBrain is a machine learning technique, and it was already the 
third-most important component of the overall search service, [lxxx] It is applied to the 
15% of searches which comprise words or phrases that have not been encountered 
before, and converts the language into mathematical entities called vectors, which 
computers can analyse directly. Microsoft also uses machine learning techniques in its 
Bing search engine. 

In February 2016 Google announced an important change of leadership in its search 
division: Amit Singhal was replaced by John Giannandrea. [lxxxi] Singhal had overseen 
the introduction of RankBrain, but was seen as having a bias against applying machine 
learning techniques to search because it is often impossible to know how the machine 
has reached its conclusions. Giannandrea has no such reservations: in his previous role 
he oversaw Google's entire artificial intelligence research activity, including deep 
learning. This succession is perhaps an allegory of the way that AI is taking over the 
internet, on the way to taking over everything else. 

One of the benefits Google hopes to obtain by increasing its use of AI in search is extra 
ammunition in its competition with Amazon. Google's competitors in search are not 
Microsoft's Bing, and certainly not Yahoo. 39% of purchases made online begin at 
Amazon, compared with 11% at Google. [lxxxii] Improving that ratio is a key aim for 

the search giant. We have seen before with the relative decline of seemingly invincible 
goliaths like IBM and Microsoft how fierce and fast-moving the competition is within 
the technology industry. This is one of the dynamics which is pushing AI forward so 
fast and so unstoppably. 

Image and speech recognition 

Deep learning has accelerated progress at tasks like image recognition, facial 
recognition, natural speech recognition and machine translation faster than anyone 
expected. In 2012, Google announced that an assembly of 16,000 processors looking at 
10 million YouTube videos had identified - without being prompted - a particular class 
of objects. We call them cats, [lxxxiii] Two years later, Microsoft researchers 
announced that their system - called Adam - could distinguish between the two breeds 
of corgi dogs. [lxxxiv] (Queen Elizabeth is famously fond of corgis, so Adam's skill 
would be invaluable in certain British social circles.) 

In February 2015, Microsoft announced that its AI systems could identify an image 
better than humans according to the tests laid down by ImageNet, the world's top image- 
recognition competition, [lxxxv] A few days later, Google announced it had done even 
better, [lxxxvi] Not to be left out, Facebook posted an impressive demonstration video 
in November 2015 . [lxxxvii] 

We humans are very good at recognising each other's faces. Throughout history it has 
been vitally important to distinguish between members of your own group who will help 
you, and members of rival groups who may try to kill you. A Facebook AI system 
called DeepFace reached human-level ability to recognise human faces in March 2014, 
scoring 97% in a test based on a database of celebrity photos called Fabeled Faces in 
the Wild fFFW). [lxxxviii] The following year it announced the ability to recognise 
faces even when they are not looking towards the camera, with 83% reliability. Google 
now offers the same functionality to users of Google +. [lxxxix] 

These tech giants are sensitive to the privacy concerns that this raises, and are throttling 
back their offerings so as not to raise alarm. But of course there is no stopping the 
progress: the genie is well and truly out of the bottle. 

In January 2016 Baidu (often described as China's Google) showed off a system called 
DuFight which uses a camera to capture an image of something in front of you, sends the 
image to an app on your smartphone, which identifies the object and announces what it 
is. One application of this is to help blind people know what they are “looking” at.jxc] 

You can download a similar app called Aipoly for free at iTunes. jxci] 

Speech recognition systems that exceed human performance will be available in your 
smartphone soon. [xcii] Microsoft-owned Skype introduced real-time machine 
translation in March 2014: it is not yet perfect, but it is improving all the time. 

Microsoft CEO Satya Nadella revealed an intriguing discovery which he called transfer 
learning: “If you teach it English, it learns English,” he said. “Then you teach it 
Mandarin: it learns Mandarin, but it also becomes better at English, and quite frankly 
none of us know exactly why.” [xciii] 

In December 2015, Baidu announced that its speech recognition system Deep Speech 2 
performed better than humans with short phrases out of context, [xciv] It uses deep 
learning techniques to recognise Mandarin. 

Learning and innovating 

It can no longer be said that machines do not learn, or that they cannot invent. In 
December 2013, DeepMind demonstrated an AI system which used a deep learning 
technique called unsupervised learning to teach itself to play old-style Atari video 
games like Breakout and Pong. jxcv] These are games which previous AI systems found 
hard to play because they involve hand-to-eye co-ordination. 

The system was not given instructions for how to play the games well, or even told the 
rules and purpose of the games: it was simply rewarded when it played well and not 
rewarded when it played less well. As the writer Kevin Kelly noted, “they didn't teach 
it how to play video games, but how to learn to play the games. This is a profound 
difference.” [xcvi] 

The system's first attempt at each game was feeble, but by playing continuously for 24 
hours or so it worked out - through trial and error - the subtleties in the gameplay and 
scoring system, and played the games better than the best human player. 

The DeepMind system showed true general learning ability. On seeing the 
demonstration, Google acquired the company for a reported $400m. 

Emulating and predicting human cognitive abilities 

In June 2015, pictures produced by Google's image recognition neural network fired the 
public imagination because of their surreal, hallucinogenic properties. In December 

2015 a group of AI researchers at MIT published a paper about a system which was 
able to predict the memorability of images better than humans. The system, called 
MemNet, reviewed a dataset of 60,000 images, and classified them in 1,000 different 
ways. It was able to identify why certain images were more memorable than others. 

To repeat, these systems are not conscious, and have no imagination. They are neither 
creating art nor being emotionally affected by the images they process. But that does not 
matter. They can analyse and process the images in ways that are important to us. And 
they can do it better than we can. 

Is it just Google? 

Google, Facebook and the other tech giants pioneered the use of machine learning, and 
for a while they were pretty much the only organisation with the expertise, the 
computing resources and the data to implement it. The joke was that machine learning 
was like teenage sex - everyone talked about it but pretty much no-one did it. That is 

Remember spam? In the late 2000s there was talk of it crashing the internet but now you 
rarely see it unless you look at your junk mail box. It was tamed by machine learning. 
The same is happening with user-generated content (UGC). We like to read the 
comments on news sites: many of them are dumb, but many are smart and funny. After 
all, there wouldn’t be any point in crowd sourcing if the crowd was all stupid. But 
some of it needs a grammatical dry-clean to be useful, and the good stuff needs to be 
surfaced. This is increasingly being done with machine learning, and by companies far 
down the pyramid from the tech giants. Companies large and small are using machine 
learning to work out what information to present to their customers and targets at every 
encounter, [xcvii] 

IBM says that its cognitive computing business, which depends heavily on machine 
learning, now accounts for over a third of its $81 billion annual revenues, and is the 
main focus for the company’s growth. IBM Watson’s best-known work today is in the 
medical sector, but it is also carrying out large-scale projects in food safety with Mars, 
and in personality profiling for recruitment firms and dating apps. [xcviii] 

In December 2015, Elon Musk and Sam Altman, president of the technology incubator Y 
Combinator announced the formation of a new company called Open AI. They had 
recruited a clutch of the top machine learning professionals despite the efforts of Google 
and Facebook to hang onto them with eye-watering financial offers. There is some 

uncertainty about whether other companies controlled by Musk and Altman (like Tesla 
and Solar City) will have privileged access to technologies developed at Open AI, but 
the thrust of the company is to make advanced AI techniques more widely available in 
the hope that will de-risk them. [xcix] 

Because it works, the use of machine learning will continue to grow - fast. 


In the last chapter we saw Robin Hanson's cynicism about “all the impressive 
computing demos [Martin Ford] has seen lately,” and Hanson's conclusion that Ford's 
“reading of the omens seems ... little better than fortune telling with entrails or tarot 
cards.” How impressed you are by demos is of course a personal matter. I side with 
Ford on this one. But a more important point is that we are still in the early days of AI 
development. Its rate of improvement is rapid, and what we will see in the next few 
years and decades will be startling. 

3.5 - Exponential future 

Big investments, different approaches 

The science of artificial intelligence is advancing rapidly, with significant steps 
announced almost every month. Enormous resources are being devoted to achieving 
these advances. 

Some of the cutting edge work in AI goes on in universities, but much of it happens 
inside the tech giants on the US West Coast. Four of them - Intel, Microsoft, Google 
and Amazon - are among the world's top ten R&D (research and development) 
spenders, with a combined budget in 2015 of $42bn.[c] This equals the entire R&D 
spend of the UK, both public and private, [ci] IBM, Apple and Facebook are not far 
behind, and are increasing their R&D spend sharply. 

In addition there are around 1,000 startup companies basing their products and services 
on AI. [cii] But despite all this, it is still early days for the sector. By one count there 
were over 300 venture capital deals in Al-based companies during 2015, but 80% of 
them were for less than $5m, and 75% of them were in the US. jciii] 


In September 2015, Google announced an important change in strategy. Having built a 
very lucrative online advertising business based on algorithms and hardware which 
produced better search results than anyone else, it was open sourcing its current best AI 
software - a deep learning engine called Tensor Flow, [civ] The software was initially 
licensed for single machines only, so even very well resourced organisations weren’t 
able to replicate the functionality that Google enjoys, but the move was significant. In 
April 2016 that restriction was lifted, [cv] 

In October 2015, Facebook announced that it would follow suit by open sourcing the 
designs for Big Sur, the server which runs the company's latest AI algorithms . [cvi] 

Then in May 2016 Google open sourced a natural language processing programme 
playfully called Parsey McParseFace, and SyntaxNet, an associated software toolkit. 
Google claims that in the kinds of sentences it can be used with, Parsey’s accuracy is 
94%, almost as good as the 95% score achieved by human linguists. [cvii] 

Open sourcing confers a number of advantages. One is a level of goodwill among the 

AI community. More importantly, researchers in academia and elsewhere will learn the 
systems, and be able to work closely with Google and Facebook- and indeed be hired 
by them. Also, having more smart people working with their systems means there are 
more smart people making suggestions about improvements and de-bugging. 

So far, Apple, the world's largest technology company (and indeed at the time of writing 
the world's largest company by stock market capitalisation) is an exception to this trend 
towards open sourcing. There are signs that this is changing, and it may have to: people 
with academic training are generally more comfortable working in organisations that 
share their findings, and many of the best people with deep learning experience have 
significant academic backgrounds. 

Extending connectivity 

Another example of the enlightened self-interest of the technology giants is Google's and 
Facebook's initiatives to extend internet access from the current level of 3.2bn people to 
the remaining 57% of the world's population. Google is experimenting with fleets of 
helium balloons which are manoeuvred around in the stratosphere (from around 11 
miles up, higher than airplanes), providing wi-fi connectivity to mobile devices below. 
In June 2015, Google signed a deal with the Sri Lankan government to make that country 
the first in the world to receive blanket wi-fi coverage, [cviii] 

The following month, Facebook revealed test flights by scale models of a 42-meter 
wingspan drone which will beam internet connectivity from the stratosphere to special 
receivers with lasers and radio. In October 2015 the two companies announced that 
their respective teams are collaborating. 


Doubling up 

If you think of artificial intelligence as a car, algorithms are the engine control system, 
big data is the fuel, and computing power is the engine. (Big data is a term coined in the 
mid-1990s by John Mashey at Silicon Graphics, a computer firm, to describe very large 
and growing data sets which could yield surprising insights into a wide range of 
phenomena. [cix] l The engine is getting more powerful at an exponential rate: its 
performance is doubling repeatedly. It is impossible to understand the scale of change 
that we face in the coming years without comprehending the astonishing impact of 
exponential increase. 

Imagine that you stand up and take 30 paces forward. You would travel around 30 
yards (or metres, if you are outside Britain and its former colonies). Now imagine that 
you take 30 exponential paces, doubling the length each time. Your first pace is one 
metre, your second is two metres, your third is four metres, your fourth pace is eight 
metres, and so on. 

How far do you think you would travel in 30 paces? The answer is, to the moon. In 
fact, to be precise, the 29 th pace would take you to the moon; the 30 th pace would bring 
you all the way back. 

That example illustrates not just the power of exponential increase, but also the fact that 
it is deceptive and back-loaded. Here is another illustration of that. Imagine that you 
are in a football stadium (either soccer or American football will do) which has been 
sealed to make it water-proof. The referee places a single drop of water in the middle 
of the pitch. One minute later she places two drops there. Another minute later, four 
drops, and so on. How long do you think it would take to fill the stadium with water? 
The answer is 49 minutes. But what is really surprising - and disturbing - is that after 
45 minutes, the stadium is just 7% full. The people in the back seats are looking down 
and pointing out to each other that something significant is happening. Four minutes 
later they have drowned. 

The fact that exponential growth is back-loaded helps explain another phenomenon, 
known as Amara’s Law, after the scientist Roy Amara. This states that we tend to over¬ 
estimate the effect of a technology in the short run and under-estimate the effect in the 
long run, [ex] 

People often talk about the “knee” of an exponential curve, the point at which past 
progress seems sluggish, and projected future growth looks dramatic. This is a 
misapprehension. When you compare exponential curves plotted for ten and 100 
periods of the same growth, they look pretty much the same. In other words, wherever 
you are on the curve, the past always looks horizontal and the future always looks 

The author JohnLanchester describes how in 1996 the US government started building a 
new supercomputer to model the behaviour of nuclear explosions. The result was Red, 
the first machine to process more than a trillion floating point operations per second (a 
teraflop). It remained the world's fastest supercomputer until 2000, but by 2006 that 
level of processing was available to schoolchildren in the SonyPS3 gaming 
computer. [cxi] This is Moore's Law at work. 

Moore's Law 

In 1965, Gordon Moore was working for Fairchild Semiconductors when he published 
a paper observing that the number of transistors being placed on a chip was doubling 
every year. He forecast that this would continue for a decade, which his 
contemporaries considered extremely adventurous. In 1975 he adjusted the period to 
two years, and shortly afterwards a Caltech professor named Carver Mead coined the 
term Moore's Law. In 1968 Moore co-founded Intel, and following an observation by 
Intel executive David House that the performance of individual transistors was also 
improving, the Law is generally taken to mean that the processing power of $1,000 of 
computer doubles every 18 months. 

Moore's Law is of course not a law, but an observation which became a self-fulfilling 
prophecy - a target and a planning guide for the semiconductor industry, and for Intel in 
particular. Moore's Law celebrated its 50 th anniversary in April 2015. Given its 
importance in human affairs there was remarkably little fanfare. 

Exponential curves do not generally last for long: they are just too powerful. In most 
contexts, fast-growing phenomena start off slowly, pick up speed to an exponential rate, 
and then after a few periods they tail off to form an S-shaped curve. However 
exponentials can continue for many steps, and in fact each of us is one of them. You are 
composed of around 27 trillion cells, which were created by fission, or division - an 
exponential process. It required 46 steps of fission to create all of your cells. Moore's 
Law, by comparison, has had 33 steps in the 50 years of its existence. 

It would take another two decades for Moore's Law to run through the same number of 
steps as human cell division. By that time, if the Law holds, the total amount of 
computing power now available to Google will be available on an ordinary desktop 
computer. Imagine what we could achieve when every teenager has Google's computing 
power in her bedroom - and try to imagine how much power Google will have by then! 

A number of other technological developments have been observed to progress at an 
exponential rate, including memory capacity, LEDs (which follow Haitz's Law [cxii] ). 
sensors (where the cost per observation is decreasing exponentiall y[cxiii] I. and the 
number of pixels in digital cameras, [cxiv] 

No more Moore? 

In 2015, Intel seemed uncertain about whether its own chip development will keep 

Moore's Law on track. This is important, as Intel (whose name is a conjunction of 
integrated electronics) has been the world's biggest chip manufacturer since 1991, and 
is also the world's third-largest R&D spending company (after Volkswagen and 
Samsung). It has led the miniaturisation of microchips. 

In February 2015 Intel updated journalists on their chip programme for the next few 
years, and the schedule maintained Moore's Law's exponential growth. [cxv] The first 
chips based on its new 10 nanometre manufacturing process were due to be released in 
late 2016 / early 2017, after which Intel expected to move away from silicon, probably 
towards a III-V semiconductor such as indium gallium arsenide, [cxvi] (That 10 
nanometres is the distance between the two nearest repeating features on the chip.) 

But in July 2015, Intel CEO Brian Krzanich said that it was taking longer for the firm to 
cut the size of its transistors: “our cadence today is closer to 2.5 years than to 2.” At the 
time of writing, the firm's smallest transistors are the 14 nanometer Skylake model, and 
the next size down will be the 10 nanometer Cannonlake, due in late 2017, a six-month 

Since 2007, Intel had pursued the development of its chips with a “tick-tock” cadence. 
The tick represented improvements in the manufacturing process, which enabled chip 
size to be reduced from 45nm to 32nm to 22nm to 14nm. The tock was improvements in 
the architecture. The new cadence announced by Krzanich was described by one 
observer as a move from tick-tock to tic-tac-toe, representing process (tic), architecture 
(tac), and optimisation and efficiency ftoe). [cxvii] 

However, Mark Bohr, a 37-year Intel veteran and senior fellow in the its processor 
technology team, argued that taking a longer view, Moore's Law still applied. He is 
working on the technology to get down to 5 nanometers, [cxviii] 

(To provide some context, a human hair is 100,000 nanometers thick, so each hair on 
your head is 10,000 times thicker than Intel's next release of transistors. Silicon atoms 
are around 0.2 nanometers across, so a 5nm structure is about 20 atoms wide.) 

Of course Intel is not the only game in town. In July 2015, IBM announced that it had a 
prototype chip of 7 nanometers, using silicon-germanium for some components. [cxix] 
Manufacturing at scale is very different from prototyping, however, and IBM did not 
expect to manufacture at scale for two more years. But this would be ahead of Intel's 
schedule, and the IBM announcement was especially important in heralding a successful 
move from Deep Ultraviolet (DUV) to Extreme Ultraviolet (EUV) lithography, which 
operates at much shorter wavelengths. 

3D chips and new architectures 

Moore's Law has undergone substantial transitions before. Until 2004, regular 
increases in the clock speeds of computer chips contributed a large part of their 
performance improvements. (See here [cxx] for an explanation of clock speeds, if you 
like that kind of thing.) Over-heating put a stop to this, and instead, chip manufacturers 
incorporated more processors, or “cores”. Modern smartphones may have four cores or 
even eight, which means the processes they work on have to be broken down into pieces 
which are operated on in parallel. 

However successfully the chip manufacturers prolong it, the existing architecture will 
reach its end point eventually. Researchers are hard at work on a number of 
technologies that could replace it. One of these is 3D chips. Placing chips side-by-side 
delays the signals between them and causes bottlenecks as too many signals try to use 
the same pathways. These problems can be eased if you place the chips on top of each 
other, but this raises new problems. Silicon chips are fabricated at 1,800 degrees 
Fahrenheit, so if you manufacture one chip on top of another you will fry the one below. 
If you fabricate them separately and then place one on top of the other, you have to 
connect them with thousands of tiny wires. 

In December 2015, researchers from Stanford announced a new method of stacking 
chips which they called Nano-Engineered Computing Systems Technology, or N3XT. 
They claimed this was a thousand times more efficient than conventional chip 
configurations. [cxxi] They did not give an estimate when mass production of N3XT 
chips might begin. 

Another approach is to combine the memory chips with the traditionally separate 
processing chips, to reduce the amount of traffic between those two. Another is to 
design chips specifically to implement neural networks, which is the approach taken by 
an MIT team that announced the Eyeriss chip in February 2016. [cxxii] 

In March 2016, scientists from IBM’s TJ Watson Research Centre announced their 
belief that “resistive processing units” which combine CPU and memory on the same 
chip could accelerate the processing of machine learning algorithms as much as 30,000 
times. [cxxiii] The following month the chip maker Nvidia unveiled its Tesla PI00 
GPU, optimised for machine learning and boasting huge performance gains. [cxxiv] And 
if you would like to know what on earth CPUs and GPUs are, go here. [cxxv] 

Towards quantum computing 

Another much talked-about route to more powerful machines is quantum computing. 

This is based on the idea that while classical computers use bits (binary digits) which 
are either on or off, quantum bits (qubits) can be both on and off at the same time - 
known as superposition. This enables them to carry out a number of different 
calculations at once. 

Google bought a quantum computer from Canadian company D-Wave in 2013, but was 
unable to demonstrate to everyone's satisfaction that it actually worked. This changed in 
December 2015, when Google's engineering director Hartmut Neven announced that its 
D-Wave computer was 100 million times faster than a traditional desktop computer in a 
"carefully crafted proof-of-concept problem". 

Keeping qubits stable is very hard, but Google thinks it is getting close, [cxxvi] IBM 
and Microsoft are also bullish about their quantum computing projects, [cxxvii] If they 
are successful, machine learning will no longer require the massive data sets and 
extensive training which are necessary today, and computers will edge that bit closer to 
human-level capabilities. 

So what? Where's this all heading? 

Moore's Law was an observation which became a target-generator rather than being a 
description of a fundamental property of the world. Keeping it on track has involved 
numerous ingenious and unpredictable steps in the past. Commercial imperatives and 
sheer human inventiveness have managed it so far, and there are plenty of avenues being 
explored which could maintain that performance. 

Using the specific definition arrived at by Gordon Moore and David House in the late 
1960s, Moore's Law ceased to hold some years ago. But as Intel's Shekhar Borkar 
observes, the meaning of Moore's Law as far as users are concerned is broader - 
namely, that the power of $ 1,000-worth of computer doubles every couple of years. 
There are plenty of ways to keep that going, and plenty of incentives too. [cxxviii] 

The people who are actually working on the technologies seem determined to maintain 
the pace of the advance. Moore is more, and more is better. Even the worst case 
predictions envisage continued rapid improvement in computer processing power, 
albeit perhaps slower than previously. 

In December 2015, Microsoft's chief speech scientist Xuedong Huang noted that speech 
recognition has improved 20% a year consistently for the last 20 years. He predicted 
that computers would be as good as humans at understanding human speech within five 

Geoff Hinton - the man whose team won the landmark 2012 ImageNet competition - 
went further. In May 2015 he said that he expects machines to demonstrate common 
sense within a decade. 

Common sense can be described as having a mental model of the world which allows 
you to predict what will happen if certain actions are taken. Professor Murray 
Shanahan of Imperial College uses the example of throwing a chair from a stage into an 
audience: humans would understand that members of the audience would throw up their 
hands to protect themselves, but some damage would probably be caused, and certainly 
some upset. A machine without common sense would have very little idea of what 
would happen. 

Facebookhas declared its ambition to make Hinton’s prediction come true. To this end, 
it established a basic research unit in 2013 called Facebook Artificial Intelligence 
Research (FAIR) with 50 employees, separate from the 100 people in its Applied 
Machine Learning team. [cxxix] 

So within a decade, machines are likely to be better than humans at recognising faces 
and other images, better at understanding and responding to human speech, and may 
even be possessed of common sense. And they will be getting faster and cheaper all the 
time. It is hard to believe that this will not have a profound impact on the job market. 

3.6 - What people do 

Jobs and tasks 

As we saw in chapter 3.2, consultants from McKinsey pointed out that machines often 
don't acquire the ability to automate entire jobs in one fell swoop. Instead they become 
able to automate certain of the tasks which people in those jobs perform. So what are 
these tasks? What exactly is it that people do for a living? 

The economies in the developed world are dominated by services, like finance, health, 
education, entertainment, retail, transport and so on. In the UK, for instance, service 
industries account for 78% of GDP, with manufacturing accounting for 15%, 
construction for 6%, and agriculture less than l%. [cxxx] 

Processing information 

In service industries, most tasks involve information: obtaining it, processing it, and 
passing it on to others. This is also true for many tasks in the manufacturing, 
construction and agricultural sectors. Obtaining information can involve carrying out 
research, asking colleagues, looking online or occasionally in books, or coming up with 
an original idea - which itself usually involves combining two or more ideas from 

Processing information can mean checking its accuracy or relevance, determining its 
importance relative to other pieces of information, making a decision about it or 
performing some kind of calculation on it. Passing information on is increasingly 
achieved electronically, for instance by email or online work flow systems. 

Obtaining, processing and passing on information can be solitary endeavours, or they 
can be carried out collaboratively with other people. Almost by definition, the solitary 
tasks can be carried out by a machine which possesses human-level (or above) ability 
to understand speech, recognise images, and a modicum of common sense. 

Working with people 

Collaboration with other humans is different. Mostly - at least for the time being. It can 
take many forms: brainstorming with colleagues; preparing for and negotiating a deal 

which will yield benefit to both sides but maximise your own; pitching an idea to a self- 
important, unimaginative and prickly boss; coaching a subordinate who has talent, but is 
also naive. These appear to be tasks which would be far harder for a machine to 

And indeed they are, but probably not for long. Even now, plenty of interactions with 
humans can be successfully automated. People seem to prefer withdrawing cash from 
ATMs than dealing with human cashiers. The centre of gravity of the entire retail 
industry is shifting online, where consumers generally avoid dealing with humans. 

This does not mean that humans are becoming anti-social - far from it. Merely that we 
like to be able to choose for ourselves when we interact in a leisurely manner with 
another human, and when we transact some business quickly and efficiently. 

Machines are sometimes surprisingly good at tasks which appear at first sight to require 
a human touch. In chapter 3.10 we will meet Ellie, a machine therapy system developed 
by DARPA, the research arm of the US military, which has proved surprisingly 
effective at diagnosing soldiers with post-traumatic stress disorder. 

Manual tasks 

We noted before that getting machines to do things that we find hard (like playing chess 
at grandmaster standard) is relatively easy, and getting them to do things that we find 
easy (like opening a door) is hard. Vivid proof of this was provided by the final round 
of the DARPA Robotics Challenge, held in June 2015. 25 robots attempted a series of 
tasks inspired by the rescue missions at the Fukushima nuclear power plant in 2011. 
None of the robots completed all the tasks, and there was a great deal of hesitation and 
falling over. 

Many jobs involving manual dexterity or the ability to traverse un-mapped territory are 
currently hard to automate. But as we will see in the next section, that is changing fast. 

Tipping points and exponentials 

New technologies sometimes lurk for years or even decades before they are widely 
adopted. 3D printing (also known as additive manufacturin g[cxxxi] 1 has been around 
since the early 1980s but is only now coming to general attention. Fax machines, 
surprisingly, were first patented in 1843, some 33 years before the invention of the 
telephone . [cxxxii] 

Sometimes the delay happens because there is at first no obvious application for the 
inventions or discoveries. Sometimes it is because they are initially too expensive, and 
engineers have to work on reducing their cost before they can become popular. And 
sometimes it is because they are simply not good enough when they are first 
demonstrated by researchers. And sometimes, of course, it is a combination of these 

Once it satisfies these conditions, a new technology can take off dramatically, with 
exciting applications which appear to most people to come from nowhere, when in fact 
the underlying technology has been known about for a long time. 

The applications of deep learning will probably be like that. The technique is a 
descendant of neural networks, which were first explored in the early days of AI in the 
mid 20 th century. Faster computers, the availability of large data sets, and the 
persistence of pioneering researchers have finally rendered them effective this decade, 
leading to “all the impressive computing demos” referred to by Robin Hanson in 
chapter 3.3, along with some early applications. 

But the major applications are still waiting in the wings, poised to take the stage. It 
won't be long now before machines are decisively better than humans at reading, 
listening, recognising faces and other images, understanding and processing natural 
language. And they won't stop at being slightly better than us. They will continue to 
improve at an exponential rate, or close to it. To say that the impact will be dramatic is 
an understatement. 

Another thing to bear in mind is that to reach the point where technological 
unemployment forces dramatic change in the way we run our economies does not 
require everyone to be unemployed and unemployable. It does not even require a 
majority to find themselves in that predicament. It just requires a substantial minority to 
believe that they will be. 

Before we proceed to look at some examples of how AI will sweep away many of the 
jobs we take for granted today, we need to quickly review some of the related 
technologies which will influence the way that happens. 

3.7 - Related technologies 

One ring to bind them 

Artificial intelligence is increasingly our most powerful technology, and it will 
increasingly inform and shape everything we do. Its full-blooded arrival coincides with 
the take-off of a series of other technologies. They are all driven at least in part by AI, 
and they will all impact the way our societies evolve. 

Because they will all unfold in different ways and at different speeds, it is impossible to 
predict exactly what the impact of these interlacing technologies will be, other than that 
it will be profound. 

The Internet of Things 

The Internet of Things (IoT) has been talked about for years - the term was coined by 
British entrepreneur Kevin Ashby back in 1 999. [cxxxiii] Indeed it has been around for 
long enough to have acquired a selection of synonyms. GE calls it the Industrial 
Internet, Cisco calls it the Internet of Everything, and IBM calls it Smarter Planet. The 
German government calls it the Industry 4.0 [cxxxiv] , the other three being the 
introduction of steam, electricity, and digital technology. As noted in chapter 2.2,1 
think this is an unhelpful term, as it shifts the IoT from the information revolution to the 
industrial one, and it under-states the importance of the information revolution. 

My favourite alternative name for the IoT is Ambient Intelligence. [cxxxv] which comes 
nearest to capturing the essence of the idea, which is that so many sensors, chips and 
transmitters are embedded in objects around us that our environment becomes intelligent 
- or at least, intelligible. 

When originally conceived, the IoT was based on Radio Frequency Identification tags 
(RFID), tiny devices about the size of a grain of rice which can be “read” remotely 
without being visible to the device which “reads” them The RFID is a passive device, 
and this concept does not involve any AI. 

Eater, technologies like Near Field Communication (NFC) were developed, which 
allow for two-way data exchange. Android phones have been NFC-enabled since 
2011, and it powers the Apple Pay system which was launched with the iPhone 6. 

The IoT is becoming possible because the component parts (sensors, chips, transmitters, 
batteries) are becoming cheaper and smaller at - yes - an exponential rate. The 

technology research company Gartner forecast in December 2013 that 26 billion 
digitally accessible devices would be installed by 2020, a 30-fold increase within a 
decade, [cxxxvi] Many of these devices have multiple sensors - smartphones can have 
as many as 30 each. [cxxxvii] 

Looking further ahead, the internet entrepreneur Marc Andreessen predicts that by 2035, 
every physical item will have a chip implanted in it. "The end state is fairly obvious - 
every light, every doorknob will be connected to the internet.” [cxxxviii] 

Making the environment intelligible offers tremendous opportunities. A bridge, 
building, plane, car or refrigerator with embedded sensors can let you know when a key 
component is about to fail, enabling it to be replaced safely without the loss of 
convenience, money, or life which unforeseen failure might have caused. This is known 
as condition-based maintenance, or predictive maintenance, and is being pioneered with 
encouraging results by MTR Corporation which runs Hong Kong’s urban transit 
network, [cxxxix] 

The IoT will improve energy efficiency across the economy, as the heating or cooling of 
buildings and vehicles can be regulated according to their precise temperature, 
humidity, etc., and the number and needs of the people and equipment using them. 

Since its launch in 1990. [cxl] the world-wide web has rendered our lives immeasurably 
easier, by placing information at our fingertips. The IoT will take that process an 
important stage further, by dramatically improving the amount and quality of 
information, and enabling us to control many aspects of our environment. You will be 
able to find out instantaneously the location and price of any item you want to buy. You 
will know the whereabouts and welfare of all your friends and family - assuming they 
don't mind - and the location of all your property: no more lost keys! You will be able 
to control at a distance the temperature, the volume, the location of things that you own. 
Your own health indicators can be made available to anyone you choose, which will 
certainly save many lives. 

Like any powerful technology, the IoT will raise concerns, particularly about privacy 
and security, and we will return to these later. It will also need a set of standards, so 
that all those semi-intelligent chairs and cars talk the same language. This may come 
about through government regulation, industry co-operation, or because one player 
becomes strong enough to impose its standards on everyone else. 

Digital assistants 

Siri, the digital assistant bundled with more recent releases of the iPhone, is a bit of a 
joke today, but by 2025 its descendants will be our constant companions, and we will 
wonder how we ever got along without them. They will be our gateway to the internet, 
and our invaluable assistants as we navigate our way through the world. 

The competition to provide the most useful digital assistant is warm and getting 
warmer. Siri was the first entrant, but many people think Google's equivalent for 
android phones is currently better. Microsoft has Cortana, and Amazon has Echo, 
which operates from an always-on fixed location device rather than a mobile. 

Facebook is betting on a mixture of AI and human intelligence with its contender, M. 
There are also numerous smaller players, of which perhaps the most interesting is Viv 
(from the Latin for “life”), a system developed by the original creators of Siri. [cxli] 
They span Siri out of a DARP A-funded research project, taking the name from Sigrid, a 
Scandinavian word meaning both “victory” and “beauty”, and sold it to Steve Jobs in 

Artificial intelligences will govern most things in our environment, and something like 
Siri will be our intermediary, negotiating with and filtering out most of the Internet of 
Things. Although we may not notice it, this will be a blessed relief. Imagine having to 
negotiate a world where every Al-enabled device has direct access to you, with every 
chair and handrail pitching their virtues to you, and every shop screaming at you to buy 
something. This dystopia was captured in the famous shopping mall scene in the 2002 
film “Minority Report”, and more laconically in Douglas Adam's peerless “Hitchhiker's 
Guide to the Galaxy” series, where the Corporation that produces the eponymous guide 
has installed talking lifts, known as happy vertical people transporters. They are 
extremely irritating. 


What generic name will be adopted for these assistants? Most of the essential tools 
which we use every day have one-syllable names, like phone, car, boat, bike, plane, 
chair, stove, fridge, bed, gu n[cxlii] . Those which have two syllables are often elided or 
rhyming, like iron and hi-fi. A few, like hoover, are named after the person or company 
who made the first successful version. 

As yet we have no short name for our digital assistants. “Digital personal assistant” and 
“virtual personal assistant” both capture the meaning but are hopelessly unwieldy. 
Maybe we'll initialise them, like TVs, and call themDAs, DP As, or VP As. Or maybe 
we'll use the brand name of one of the early leaders, and call them all Siris. Google's 

chairman Eric Schmidt came up with the interesting idea that we'll find ways to name 
them after ourselves, and his would be called “not-Eric”. [cxliii] Perhaps - and this is 
my favourite - we'll just call them our “Friends”. 

Wearables, insideables 

At the moment, the vessel which transports the primitive forebears of these essential 
guides is the smartphone, but that is merely a temporary embodiment. We will surely 
progress from portables to wearables (Apple Watch, Google Glass, smart contact 
lenses...) and eventually to “insideables”: sophisticated chips that we carry around 
inside our bodies. 

You doubt that Google Glass will make a comeback? The value of a head-up display, 
where the information you want is displayed in your normal field of vision, is 
enormous; that's why the US military is happy to pay half a million dollars for each 
head-up display helmet used by its fighter aircraft pilots. 

Apple Watch has been successful because some people will pay good money to simply 
raise their wrist rather than go to all the bother of pulling their smartphone out of their 
pocket. How much better to have that hunger for the latest bit of gossip sated, and that 
essential flow of information about your environment displayed right in front of your 
eyes with no effort whatsoever? 

With regard to “inside-ables”, the technology to enable a chip implanted inside you to 
project imagery into your field of vision is far ahead of where we are now. But it is the 
next logical step in the process after wearables, and with key aspects of technology 
advancing at an exponential rate it would be foolish to write it off. 

Screens will be everywhere by this time, of course: on tables, walls both interior and 
exterior, on the backs of lorries so that you can see what is ahead of them. [cxliv] But 
we will want to carry our own screens around with us, not least because we won't 
always want other people to see what we are looking at. 

We will probably also need to invent a new type of interface to enable us to 
communicate with our digital assistants. The 2013 movie “Her” is one of Hollywood's 
most intelligent treatments of advanced artificial intelligence. (I realise that isn't saying 
much, but Hollywood does frame the way many of us think about future technologies.) 
The essence of the plot is that the hero falls in love with his digital assistant, with 
intriguing consequences. Although he uses keyboards occasionally, most of the time 

they communicate verbally. 

There will be times when we want to communicate with our “triends” without making a 
sound. Portable “qwerty” keyboards will not suffice, and virtual hologram keyboards 
may take too long to arrive - and they may feel too weird to use even if and when they 
do arrive. Communication via brain-computer interfaces will take still longer to 
become feasible, so perhaps we will all have to learn a new interface - maybe a one- 
handed device looking something like an ocarina [cxlv] . 

Another way we may communicate with our Friends, and indeed with many of the newly 
intelligible objects in the Internet of Things is radar. In May 2015 Google posted a 
video to introduce Soli, a project which embeds a sophisticated radar sensor in a tiny 
chip. It uses no lenses, and there is nothing to break. It generates a virtual tool in the 
space above or in front of itself, a way to interpret human intent by tracking the tiniest 
motion of the human hand and fingers. Soli generates virtual representations of controls 
we are all familiar with, such as volume knobs, on-off buttons, and sliders, [cxlvi] 

Doing business with Friends 

Friends will be very big business, and the evolution of their industry will be 
fascinating. Will it turn out to be a natural monopoly, where the winner takes all? If so, 
the winner will find itself the subject of intense regulatory scrutiny, and probably of 
moves to break it up or take it into public ownership. Or will there be a small number 
of immensely powerful contenders, as in the smartphone platform business, where 
Apple and Android have the field almost to themselves? 

Will we all choose one brand of Friend at an early age, or during adolescence, and stick 
with it for life, as many people do with smartphones? Doubtless the platform providers 
will seek to lock us in to that kind of loyal behaviour. Or will we be promiscuous, 
hopping from one provider to the next as they jostle and elbow each other, taking turns 
to launch the latest, most sophisticated software? 


The final round of the DARPA Robotics Challenge in June 2015 could have been a 
triumphal display of engineering prowess and the potency of artificial intelligence. (For 
what are robots, but the peripherals of AI systems, just as mouses and keyboards are the 
peripherals of PCs?) Instead, as we noted above, it was a sad affair, with the winning 
machine taking nearly 45 minutes to complete a series of eight tasks that a toddler could 

accomplish in 10 minutes. Moreover, the tasks had already been scaled down from the 
initial targets set in 2012 when it became obvious that none of the teams were going to 
be able to achieve them. [cxlvii] 

But remember the progress in self-driving cars provoked by the DARPA Grand 
Challenge. In the initial event in 2004 the winning car drove just seven of the 150 miles 
of the track before crashing. A dozen years later, self-driving cars are demonstrably 
superior to human drivers in almost all circumstances, and they are closing the 
remaining gap fast. As far as robotics is concerned, we are at 2004 again. And don't 
forget the power of exponential improvement. 

In chapter 2.3 we met Baxter, a new generation of industrial robot, which is beginning 
to demonstrate that robots can be flexible, adaptable, and easy to instruct in new tasks. 
Research teams around the world are teaching robots to do intricate tasks. In October 
2015, a consortium of Japanese companies unveiled the Laundroid, a robot capable of 
folding a shirt in four minutes, [cxlviii] Meanwhile at the University of California, a 
team developing the Berkeley Robot for the Elimination of Tedious Tasks (BRETT) 
spent seven years reducing the time to fold a towel from 20 minutes to 1.5 
minutes, [cxlix] 

So robots can fold towels - slowly, but it will be a few years before they can carry out 
efficiently all the tasks that a hotel chambermaid does. What they can already do, 
however, is automate many of the individual tasks that chambermaids carry out. Half a 
dozen big-name hotels in California are experimenting with robots that deliver towels 
and other items to guests' rooms on demanded] Apparently toothpaste is the most-often 
requested item, presumably to keep all those perfect Hollywood teeth sparkling. 

In mid-2015, a team at University of California, Berkeley, announced that by applying 
deep learning to the problem, they could get robots to screw tops onto bottles and 
remove nails from wood with a claw hammer, and do so with approximately the same 
speed and dexterity of a human, [cli] 

Researchers are trying out different ways to improve robot performance. Teams at 
Carnegie Mellon University in Pittsburgh and at Google are getting robots to learn about 
their physical environment by having them simply prod, poke, grasp and push objects 
around on a table-top, in much the same way that a human child learns about the 
physical world. Having collected a large data set from this activity, the systems turn out 
to be better at recognising images from the ImageNet database than systems which have 
not had the physical training, [clii] 

Google's robot army - the dog that didn't bark 

In late 2013, Google announced the purchase of no fewer than eight robotics 
companies. (Since you ask, they are Boston Dynamics - purveyor of the famous Big 
Dog and Atlas models - Bot and Dolly, Meka, Holomni, SCHAFT, Redwood, Industrial 
Perception, and Autofuss.) Google also announced that the new division which owned 
them would be run by Andy Rubin, who created a huge global business with the 
Android phone platform. 

A year later, in October 2014, Andy Rubin left Google to found a technology startup 
incubator, which prompted observers to remark that Google had been surprisingly quiet 
about its collection of robot makers. In early 2016, rumours spread that Google was 
considering selling Boston Dynamics, the creator of Big Dog and Atlas, two of the 
world’s most impressive robots. Google is an experienced acquirer of companies - by 
the end of 2014 it had acquired 170 of them - and it expects them to make an impact. 

The hurdle for potential acquisition targets is the “toothbrush test”, meaning that their 
services must be potentially useful to most people once or twice every day. Sooner or 
later its robotic companies will impress us. 

Complicated relationships 

It is going to take us humans a while to get used to having robots around. A French 
company called Aldebaran, which is owned by Japanese firm Softbank, manufactures a 
robot called Pepper. 120Cm tall and costing around $ 1,200, they have a limited ability 
to “read” human emotions and respond appropriately. They have proved extremely 
popular in Japan, with four batches of 1,000 selling out in less than a minute when they 
went on sale in September 2015. 

The response to Pepper has not been straightforward, however. The manufacturer felt 
obliged to outlaw any attempt to engage in sex with the robot, and a Japanese man was 
prosecuted for assaulting one when drunk, [cliii] 

A robot called Hitchbot managed to cross Canada from coast to coast in 2015, but was 
attacked and decapitated in Philadelphia when it tried to repeat the performance in the 

More robots: androids, drones and exoskeletons 

It is not clear that robots need to resemble humans closely to perform their tasks, but that 

doesn't stop researchers from trying to make them. (Robots with human appearance 
used to be what the word “android” meant before Google appropriated it for phone 
software.) We are probably quite a few years away from having robots with the 
verisimilitude of the ones in the film “Ex Machina”, or the TV series “Humans”, for 
example. Nadine is a state-of-the-art prototype working as a receptionist at Nanyang 
Technological University in Singapore. It is humanoid, but doesn’t fool anyone who 
takes a second glance, [civ] Modelled on its inventor, professor Nadine Thalmann, it 
cannot walk, but it can smile, turn its head, and shake your hand. Its voice is powered 
by an AI similar to Siri. 

Most robots will probably be special-purpose devices, constructed to carry out a very 
specific task. An example is the Grillbot, a robot the size of a table tennis bat which 
cleans your barbecue grill, and is otherwise entirely useless. [clvi] 

Another form of robot which is taking off fast is drones - flying machines that can be 
controlled remotely or autonomously. They have a wide range of applications, 
including taking surreptitious photos of celebrities, taking selfies for life-logging 
Millennials, and delivering parcels for Amazon. They present a serious challenge for 
regulators concerned about the impact on more established forms of aircraft. These 
challenges cannot be dismissed or regulated away: internet-connected drones with 
powerful sensors and computers on board are quickly becoming essential tools for 
companies in the utilities and engineering industries, as well as government 
agencies, [clvii] 

Some people argue that exoskeletons are wearable robots. Whether or not that is 
semantically correct, they will certainly enable one human to do the work of several. At 
the moment, leading companies in the space like Ekso Bionics [clviii] are focusing on 
patient rehabilitation systems. But before long similar equipment will be available for 
people carrying out physically demanding tasks in the military, manufacturing, and 

Virtual Reality 

During 2014, many people got their first taste of virtual reality (VR) from Google 
Cardboard, an ingenious way to let smartphones introduce us to this extraordinary 
technology. 2016 is widely expected to be the year that VR really takes off, as 
Facebook's Oculus VR launches Rift, the first VR equipment for consumers that offers 
high definition visuals and no latency. Latency is a failure of synchronisation between 
the stimuli from different sources reaching the brain: if your visual experience is out of 
synch with your other senses, your brain gets confused and unhappy, and can make you 

feel surprisingly sick. [clix] 

When VR is effective it is surprisingly powerful. When the sense data being received 
by the brain become sufficiently realistic, the brain “flips”, and decides that the illusion 
being presented is the reality. 

Google is not giving up on smartphone-based VR. Having sold more than 5m of the 
cardboard units, it plans to launch a more robust plastic version in 2016, with better 
sensors and lenses. It will remain considerably cheaper than the Oculus Rift, which 
will cost hundreds of dollars . [clx] 

Augmented reality (AR) is similar to VR except that it is overlaid on your perception of 
the real world rather than replacing it. It can make elephants swim through the air in 
front of you, or plant a skyscraper in your back garden. This is handy if you want to 
remain alert to the threat from dogs and potholes while you are hallucinating swimming 
elephants. Microsoft's Hololens is the best-known AR brand to date, but great things 
are expected from a company called Magic Leap, in which Google has a substantial 

Insofar as there is any debate about whether VR is going to be an important 
development, it's between those who think it's going to be huge, and those who think it's 
going to change everything. Gartner expects two million VR headsets will be sold in 
2016, with the volume increasing thereafter - you guessed it - exponentially, [clxi] 
Digi-Capital, a specialist consultancy, expects VR and AR sales to reach $150bn within 
five years. It expects AR to account for 80% of that revenue . [clxii] And these are just 
the projections for the early years - Gartner expects the VR industry will take five years 
to achieve mainstream adoption, [clxiii] 

Digi-Capital's $150bn figure includes software as well as hardware; Goldman Sachs, a 
bank not given to hyperbole, expects hardware sales alone to reach $80bn in a decade. 
Interestingly, they expect less than half of that to come from gaming and films, with the 
rest coming from commercial and professional applications. [clxiv] 

Applications of VR and AR 

The biggest application in the short term is expected to be video games, which is no 
small playing field, since gaming has for some time rivalled Hollywood for leadership 
in global sales of packaged entertainment, [clxv] Judging by the content already being 
made available for Google Cardboard, people also enjoy ersatz travel, and adventurous 

experiences. VR versions of Google Street View let you wander around Manhattan 
until the latency makes you ill, and other developers offer you rollercoaster rides, and 
adventure sports from skiing to hang gliding. 

In the longer term the potential applications are bewildering. Without ever leaving our 
armchairs we may soon be able to enjoy such realistic simulations of events like sports 
matches and music concerts that many people will question the value of struggling with 
transport and crowds to attend the real thing. Of course, the crowd has a lot to do with 
making the event exciting in the first place, so the organisers of VR events will want to 
find a way to recreate the effect of being in a crowd. Except that you’ll be able to sit 
next to your friend, who happens to be in a VR rig a couple of continents away at the 

Education and informal learning is also likely to experience a VR revolution. How 
much more compelling would it be to learn about Napoleon by experiencing the battle 
of Waterloo than by reading about it, or listening to a lecturer describe it? How much 
easier would it be for a teacher to explain the molecular structure of alcohol by 
escorting her pupils round a VR model of it? 

Businesses will find many uses for VR, and because they often have larger budgets than 
consumers and educational institutions, they may sponsor the creation of the most 
cutting-edge applications. Computer-aided design environments will become startling 
places to work, for instance, allowing architects, designers and clients to explore and 
discuss buildings in great detail before ground is broken. And who knows what uses the 
military will find for VR. One frightening thought is that VR could become a powerful 
and truly terrifying instrument of torture, [clxvi] 

Telecommunication will also be taken to a new level. Although audio-only phone calls 
still predominate, good video-conferencing facilities add enormously to the 
effectiveness of a long-distance conversation, and the additional step of feeling present 
in the same space will improve the experience again. Anything which involves your re¬ 
location in time or space should be fertile ground for VR. 

On the other hand, it is not yet clear whether VR will turn out to be a good medium for 
movies. In a film, the director wants to direct your attention, and it isn't helpful if half 
the audience is busy gawping at images or events 180 degrees removed from the focus 
of the action. 

Cynics will point out that new media (TV, video, the internet were all new media in 
their early days) are established only when users have found ways to apply them to 

porn, gambling, and then sport. No doubt VR will make its contribution to these areas 
of human activity, but I'm not going to get sucked into a discussion of what could be 
achieved with haptic suits - clothing which allows users to experience sensations of 
heat and touch initiated remotely by someone else. 

The death of geography has been declared numerous times, but despite the rise of 
telephony, digitisation and globalisation, business and leisure travel just keeps on 
growing. Could the sense of genuine “presence” which good VR confers finally make 
the old chestnut come true? Will talent continue to be drawn into the world's major 
cities, or will VR puncture their inflated real estate prices, and smear humanity more 
evenly across the planet? 

Maybe virtual reality can render scarcity less valuable, and less problematic. In the 
real world, not everyone can live in a beautiful house on a palm-fringed beach, drive an 
Aston Martin, and be greeted by a Vermeer as they enter their living room. With virtual 
reality, everyone can - to a fair degree of verisimilitude. As we will see in chapter 5, 
this might turn out to be extremely important for our overall well-being as a species. 

The world-wide web has given us something like omniscience, and virtual reality looks 
set to give us something like omnipresence. Perhaps all we need now is a technology to 
give us something like omnipotence. 

Related concerns 

Powerful new technologies can produce great benefits, but they can often produce great 
harm. There are four serious concerns about the technologies we have just reviewed: 
privacy, security, isolation and inequality. 


AI runs on Big Data, and the Internet of Things will generate it by the bathful. In an 
intelligent environment, the whereabouts of every citizen is easy to establish, along with 
who they have met and very possibly what they discussed. Many people are 
understandably concerned about this information being used and mis-used by all sorts of 
organisations, including (depending on their political persuasion) governments, 
corporations, pressure groups and resourceful individuals - such as jealous spouses. 

As one group of activists puts it, the net is closing around us and we are increasingly 
transparent to organisations which are increasingly opaque to us. [clxvii] 

Law Enforcement is where we would expect to find the cutting edge uses of this 
technology. A company called Intrado provides an AI scoring system to the police in 
Fresno, California. When an emergency call names a suspect, or a house, they police 
can “score” the danger level of the person or the location and tailor their response 
accordingly, [clxviii] Optimists would say this is an excellent way to deploy scarce 
resources. Pessimists would reply that Big Brother has arrived. 

Others hope that we can retaliate against this kind of “surveillance on steroids” with 
“sousveillance”. With cameras ubiquitous - including on drones - the actions of those 
in authority are constrained because they know that their actions are observed and 
recorded by members of the public. This is already happening with law enforcement, 
with police officers in the US being prosecuted for harassment in situations where they 
would previously have been immune from oversight. Some authorities are actively 
embracing this development, with police officers being required to wear cameras at all 
times in order to pre-empt false allegations. With cameras on drones, the reach of 
civilian oversight can be extended so far that some are calling it “Little 
Brother”, [clxix] With the watchers being watched, we may arrive at a balance called 
“co-veillance”. [clxx] 

The arms race over data will continue between governments, large organisations, and 
the rest of us. Hollywood loves the trope of the socially dysfunctional hacker who is 
smarter, more up-to-date and more motivated than her opposite numbers in the civil 
service, but perhaps we should not be comforted by that idea. When forced to choose 
between privacy and the opportunity to share, we generally choose to share. We leave 
a trail of digital breadcrumbs wherever we go, both in the real world and online, and 
most of us are careless about it. 

In part this is because many of us feel that we have nothing to worry about because we 
have nothing to hide. But there is a chilling effect on free speech if we start to censor 
ourselves because we want to stay that way. We might think twice before entering a 
certain term into a search engine, or hesitate before making friends with someone who is 
overtly counter-cultural. Recent research shows that we self-censor when we are 
aware of the possibility that we are being surveilled, even when we know we are 
saying and writing nothing illegal. [clxxi] 

In 2015, the Chinese government provided a chilling demonstration of where this could 
lead. It is building a “social credit” database of all citizens which ranks them according 
to their trustworthiness. In a frightening extension of credit scoring systems, the 
database will incorporate all the financial and behavioural information the government 
can accumulate, and distil it into a single number, ranging from 350 to 950. A score 

above 600 qualifies a citizen for an instant loan worth $800. At 650 you can rent a car 
without leaving a deposit. At 700 you are fast-tracked for a Singapore travel permit. 
Important jobs will require high scores. 

Citizens will earn demerits for reprehensible shopping habits (too many video games? 
too much wine?) and merits for socially responsible actions, like reporting bad 
behaviour by others . [clxxii] A particularly scary aspect of the system is that people 
receive demerits if their triends on social media are marked down. 

The system will be compulsory for every Chinese citizen in 2020, and until then, eight 
pilots are being run by Chinese companies, including Sesame Credit, the financial wing 
of Alibaba, which is China's version of Amazon. 

The US civil liberties pressure group ACLU thinks “China’s nightmarish Citizen Scores 
are a warning for Americans. ... There are consistent gravitational pulls toward this 
kind of behavior on the part of many public and private US bureaucracies, and a very 
real danger that many of the dynamics we see in the Chinese system will emerge here 
over time.” [clxxiii] 

Big Data and AI could enable governments to build apparatus of control which would 
make Big Brother in George Orwell's “1984” look amateurish. 

You’re not necessarily safe from this prospect just because you live in a multi-party 
democracy. In April 2014, Nicole McCullough and Julia Cordray founded Peeple, an 
app which will enable people to rate each other according to their courtesy and 
helpfulness. Originally conceived as a way to improve behaviours, it was widely 
criticised as likely to become a medium for personal attack and bullying. The founders 
responded by changing the rules so that subjects would have a veto over any comments 
made about them on the site, although they left open the possibility that users who pay 
extra could see un-censored inputs . [clxxiv] 

Clearly we still have much to learn about how to conduct ourselves individually and 
collectively in the new world of data tsunamis and massive analytic horsepower. 

Researchers at Google and Microsoft are experimenting with promising approaches to 
squaring the circle of protecting privacy while sharing data. Working with Cornell 
University in New York, Google is trying to enable groups of organisations (e.g., 
hospitals) to train deep learning algorithms on their own separate data files and then 
share the outputs of the trained data. They have found that this can work almost as well 
as combining all the data into one file and using that to train the algorithm. 

Microsoft is using a technique called homomorphic encryption to perform analysis on 
data which is encrypted. It yields encrypted results which can then be decrypted 
without the sensitive data ever having been available to the analysts in unencrypted 
form [clxxv] 


In his book “Future Crimes”, security expert Marc Goodman sets out in detail how 
criminals, governments, and organisations use the swelling oceans of data being 
transmitted about us to steal from us and manipulate us. Cyber crime is probably the 
fastest-growing type of crime all over the world; much of it goes undetected, and much 
of what is detected goes unsolved. 

Another growing concern about hacking is sabotage. As the internet of things is built out 
and more and more of our vehicles, buildings and appliances rely on artificial 
intelligence, the problems that can be caused if their control systems are hacked 
increase in significance. The possibility of a hacker gaining control of every self- 
driving car in a city and making them all turn left at the same moment is frightening. 

Programmers say that there is no such thing as 100% security: IT systems are designed 
by humans, and we are fallible. They are also increasingly opaque, and hard to de-bug. 
An optimist would say that although complex, well-defended systems come under 
frequent attack, they are rarely successfully hacked. No hacker has yet launched a US 
nuclear missile, although of course that doesn't mean that it will never happen. Eternal 
vigilance is the price we must pay to avoid disaster, and we are not practising it at the 
moment. Many of us are lax about safeguarding our internet passwords, and many 
companies' security arrangements also fall far short of best practice. 

Policemen say that when they are pursuing a criminal, the criminal needs to be lucky all 
the time whereas the police only need to get lucky once. But when the criminals are on 
the offensive, looking for security gaps, the boot is on the other foot. 


Every time a new technology is launched, people worry that only the rich will have 
access to it, and there will be a “digital divide” separating the haves and the have-nots. 
The life experiences and opportunities of the wealthy will diverge unacceptably from 
those of the rest of us. 

So far, while not groundless, this fear has been exaggerated. It is true that in recent 
years the super-rich have gained more income and wealth than anyone else in most 
developed economies. (And we're not talking about the 1% here, but the 0.01%.) 
Meanwhile, there are people who struggle to afford what many would consider the 
basic necessities of life - although the definition of basic necessity varies greatly 
between developed countries and elsewhere in the world. 

It is also true that the disparity of income between average people in rich countries and 
average people in the poorest countries is enormous. This disparity, however, is 
shrinking. And those in America and Europe who protest about the obscene wealth of 
the 1% in their own countries seem curiously un-troubled by the fact that they 
themselves are often among the richest 1% of the world's population. 

The new technologies which have emerged during the various stages of the industrial 
revolution have become available to most people in developed economies not long after 
they were invented. The car, the refrigerator, the washing machine, the TV, the home 
computer, the smartphone - all have gone through the same cycle. An expensive first 
version is launched which can be afforded only by the wealthy. It doesn't work very 
well, and is at least in part a status symbol. Very quickly, the technology improves and 
the price falls, and pretty soon the great majority of us have one. Next in line for this 
cycle is virtual reality headgear. 

The reason for this is simple economics. Companies make far more money by selling 
lots of cheap smartphones (for instance) to everyone than by selling a few very 
expensive ones to the wealthy elite. And in a competitive economy, even if the first 
company to market is happy to make its money by scalping the rich, other companies 
will quickly come along to raise the quality and reduce the price. There is no “fridge 
divide”; why should there be a “digital divide”? 

I said this fear was exaggerated “so far”. In chapter 5 we will see that there may be 
more grounds for concern in the not-too-distant future. 


Parents have long fretted about their teenage children spending long hours in antisocial 
isolation, hunched over a video game console. Wishing their kids would go outside and 
kick a ball around instead, they have agonised over a series of scares about the ill 
effects of video games, which allegedly make kids violent, stop them developing social 
skills, render them vulnerable to legions of grooming molesters, and give them 

impossibly short attention spans. And the blue light of the screen disrupts their sleep. 

Meanwhile, the Flynn Effect describes the finding that IQ levels are increasing steadily 
each generation, [clxxvi] which should not be surprising when you consider the general 
trends toward less smoking, less drinking, better central heating, better food and better 
healthcare. And the fact that we are continually learning more about what works in 
education and what does not. 

Humans are intensely social creatures. The need to belong to a tribe - to be accepted 
by it and perhaps to climb its hierarchy - is programmed deeply into us. Working 
together in tribes is how we survived in the savannah, surrounded by animals which 
were stronger and faster, with bigger teeth. Individuals who were cast out of their tribe 
quickly joined another one or - more likely - got eaten. It would be amazing if in a 
single bound, one generation of teenagers suddenly freed themselves from this 
evolutionary programming and isolated themselves in solitary pursuits. 

And indeed they haven't. The most popular video games are those which people can 
play together, and incorporate into their social bonding activities. For teenagers, these 
activities are just as important as they ever were - and of course it is no less important 
that their parents be at least slightly appalled by them 

Of course, if and when the day comes when people can plug into utterly compelling 
virtual reality worlds through a direct neural link, and effectively disappear into the 
Matrix, things may be different. But unless we have altered our cognitive make-up 
dramatically by then, my hunch is that we will find a way to make the Matrix social too. 

We have explored the state of the art in artificial intelligence, and peered into its likely 
future, along with related technologies that it drives and will be affected by. We have 
considered what people do at work. Now it is time to think about the kinds of jobs 
which will be automated by these technologies. 

We’ll start with driving. 

3.8 - The poster child for technological unemployment: self-driving 


The case for introducing self-driving cars is simple and overwhelming: around the 
world, human drivers kill 1.2 million people a year, and injure a further 20 to 50 
million, [clxxvii] Road traffic accidents are the leading cause of death for people aged 
15 to 29, and they cost middle-income countries around 2% of their GDP, amounting to 
$ 100 billion a year. 

90% of these accidents are caused by human error, [clxxviii] Humans become tired, 
angry, drunk, sick, distracted or just plain inattentive. Machines don't, so they don't 
cause accidents. To paraphrase Agent Smith in “The Matrix”, we are sending humans to 
do a machine's job. 

There is also the wasted time and frustration. We all know that driving can be fun, but 
not when you're stuck in traffic - perhaps because one of your fellow humans has caused 
an accident. On average, American commuters spend the equivalent of a full working 
week stuck in traffic every year - twice that much if they are lucky enough to work in 
San Francisco or Los Angeles . [clxxix] We drive rather than use public transport 
because there is no appropriate public transport available, or sometimes because we 
prefer travelling in our own space. Self-driving cars could give us the best of both 
worlds, allowing us to read, sleep, watch video or chat as we travel. 

Finally, self-driving cars will enable us to use our environments more sensibly, 
especially our cities. Most cars spend 95% of their time parked, [clxxx] This is a waste 
of an expensive asset, and a waste of the land they occupy while sitting idle. We will 
consider later how far self-driving cars could alleviate this problem. 

To autonomy and beyond 

Self-driving cars, like our artificially intelligent digital assistants, are still waiting to 
receive their generic name. “Self-driving cars” is the name we are stuck with for the 
time being, but it is all clunk and no click. At the end of the 19 th century it was 
becoming obvious that horseless carriages were here to stay, and needed a shorter 
name. The Times newspaper adopted “autocar” but the Electrical Engineer magazine 

objected that it muddled Greek (auto) with Latin (car). It argued instead for the 
etymologically purer “motor-car”, [clxxxi] Perhaps we will contract the phrase 
“autonomous vehicle”, and call them “autos”. 

Some people are going to hate self-driving cars, whatever they are called: petrol-heads 
like Jeremy Clarkson are unlikely to be enthusiastic about the objects of their devotion 
being replaced by machines with all the romance of a horizontal elevator. Some people 
are already describing a person who has been relegated from driver to chaperone as a 
“meat puppet”, [clxxxii] 

The US Department of Transport draws a distinction between (partly) autonomous cars 
and (fully) self-driving cars. [clxxxiii] The former still have steering wheels, and 
require a human driver to take over when they encounter a tricky situation. Self-driving 
cars, by contrast, are fully independent, and the steering wheel has been removed to 
save space. Autonomous cars will probably be merely a staging post en route to the 
completely self-driving variety. 

In fact the US DoT grades cars on a scale from LO, where the driver does everything, to 
L4, where the car does everything. Google’s initial idea was that the first self-driving 
cars in general use would be L3, meaning that the human driver should be ready to take 
over at a moment’s notice if anything went wrong, just as airplane pilots are. But the 
technology proved so reliable that its test drivers became complacent and engaged in 
“silly behaviour”. For instance, one turned round to look for a laptop in the back seat 
when the car was doing 65 mph. This experience persuaded Google to advocate 
immediate adoption of L4. [clxxxiv] 

The state of the art 

Self-driving cars have come a long way since 2004, when the humvee Sandstorm got 
stuck on a rock seven miles into the first DARPA Grand Challenge, but they are not 
perfect yet. They struggle with heavy rain or snow, they can get confused by potholes or 
debris obstructing the road, and they cannot always discern between a pedestrian and a 
policeman indicating for the vehicle to stop. A self-driving car which travelled 3,400 
miles from San Francisco to New York in March 2015 did 99% of the driving itself, but 
that means it had to hand over to human occupants for 1% of the journey. [clxxxv] With 
many technology projects, resolving the last few issues is more difficult than the bulk of 
the project: edge cases are the acid test. Nevertheless, those edge cases are being 
tackled, and will be resolved. 

It is well-known that Google's self-driving cars have travelled well over a million 
miles in California without causing a significant accident, but what is less well-known 
is that the cars also drive millions of miles every day in simulators. Chris Urmson, 
head of the Google project, expects self-driving cars to be in general use by 
2020. [clxxxvi] 

Sceptics point out that Google's self-driving cars depend on detailed maps. But 
producing maps for the roads outside California doesn't sound like an insurmountable 
obstacle, and in any case, systems like SegNet from Cambridge University enable cars 
to produce maps on the flv. [clxxxvii] 

A fully autonomous bus made in France has been serving the centre of the Greek city of 
Trikala since February 2015. It travels at a top speed of 20 mph along a pre-determined 
route which is also used by pedestrians, cyclists and cars. [clxxxviii] 

In December 2015 Bloomberg reported that Google was preparing to move its self¬ 
driving cars unit from its Google X research arm to become a stand-alone business unit 
within the Alphabet holding company, [clxxxix] At the same time, Elon Musk, CEO of 
Tesla, remarked that he was revising his estimate of the time when fully automated cars 
would be available from three years down to two. [cxc] In January 2016 he announced 
that within about two years, Tesla owners would be able to “summon” their driverless 
car from New York to pick them up in Los Angeles . [cxci] He claimed that Tesla cars 
are already better drivers than humans. [cxcii] In April 2016 he went further, claiming 
that Tesla’s autopilot system was already reducing the number of accidents by 50% - 
where an accident meant an incident where an airbag was deployed, [cxciii] 

Ford reported success in January 2016 with tests of its self-driving car in snowy 
conditions. Unable to determine its location by the obscured road markings, it navigates 
by using buildings and other above-ground features. [cxciv] In May 2016 an executive 
in Ford’s autonomous vehicle team estimated that the remaining technological hurdles 
would be overcome within five years, although adoption would of course take longer. 

He said the amount of computing power each car currently required was “about the 
equivalent of five decent laptops.” [cxcv] 

At the time of writing, the only accident which a Google self-driving car might be 
blamed for happened in February 2016. The car was trying to merge into a line of 
traffic and expected that a bus which was approaching from behind would give way. It 
didn’t. The car was travelling at 2 mph and no-one was hurt, so no police report was 
filed to attribute blame officially. The bus driver has declined to comment, [cxcvi] 

Of course, just because a product becomes available, that doesn't mean it will be 
bought, still less that it will comprehensively replace the existing population of products 
that it is designed to supplant. The rate at which that happens, if it happens at all, 
depends on a host of factors including regulation, price, design, service support, 
promotion and PR, and the length of the replacement and upgrade cycle for the product 

Regulation is an important consideration. Google was disappointed when California's 
Department of Motor Vehicles (DMV) proposed new rules for self-driving cars in 
December 2015 which banned vehicles which lacked the capacity for a human to take 
control. In theory, unco-operative regulators could slow or even stop the arrival of 
self-driving cars, and there will be powerful lobbies pressing for this. But they can 
only succeed if all regulators everywhere agree, and work together, and that will not 
happen - even within the US, never mind globally. In 2015 Google expanded its test 
driving programme beyond Silicon Valley to Austin, Texas, where the authorities 
welcome the tech giant's research money and prestige, [cxcvii] In 2016 it added two 
more cities, Kirkland in Washington and Phoenix in Arizona, [cxcviii] 

Several European countries (including the UK) are keen to burnish their credentials as 
leaders in what will undoubtedly be a massive new industry. 

The impact on cities 

Enthusiasts for self-driving cars sometimes paint a utopian picture of cities where 
almost no-one owns a car because communally-owned taxis are patrolling the streets 
intelligently, anticipating our requirements and responding immediately to our 
summons. Whereas today, our cars sit idle 95% of the time, squatting like polluting 
toads on vast acres of city land, in this bright tomorrow they are used efficiently, and the 
land given over to parking can be returned to pedestrians and useful buildings. Traffic 
flows smoothly because the cars are in constant communication with each other: they 
don't bunch into jerky waves and they don't need to stop at intersections. 

This is almost certainly an exaggeration. There will still be peak times for journeys, so 
even if most journeys are undertaken in communal cars, many of them will be parked up 
during off-peak hours. And traffic will still have to halt at intersections every now and 
then if pedestrians are ever going to be able to cross the road. Not every pedestrian 
crossing can have a bridge or an underpass. 

Nevertheless, machine-driven cars will be more efficient consumers of road space than 
human drivers. Traffic conditions are not fixed fates which once imposed can never 

improve. A congestion charge has significantly reduced traffic flows in London, and the 
switch to almost-silent hybrid taxis has made walking the streets of Manhattan an even 
better experience than it used to be. [cxcix] In any case, more efficient road use is not 
required to justify the introduction of self-driving cars. The horrendous death and injury 
toll imposed by human drivers is sufficient, together with the liberation from the 
boredom and the waste of time caused by commuting. 

Detroit's response 

The car manufacturing industry first experimented with self-driving cars decades ago. 
From 1987 to 1995, the European Union spent $750m with Daimler Benz and others on 
the Prometheus project (the PROgramme for a European Traffic of Highest Efficiency 
and Unprecedented Safety). [cc] There were some impressive technical achievements, 
but ultimately the project faded. Fortunately, among other things, we have got better at 
devising acronyms since then. 

The automotive industry's response to the implicit challenge from Google and others has 
been slow and piecemeal. In part this is because the car industry thinks in seven-year 
product cycles, while the technology industry thinks in one-year cycles at most. Most of 
the large car companies seem convinced that self-driving technology will be introduced 
gradually over many years, with adaptive cruise control and assisted parking bedding in 
during the lifetime of one model, and assisted overtaking being introduced gradually 
with the next model, and so on. That is far too slow for the tech titans of Silicon 
Valley. Google, Tesla, Uber and others are racing towards full automation as soon as it 
can be safely introduced. If Detroit does not join in it may find itself displaced. 

In the closing months of 2015, Detroit and its rivals seemed to wake up. Toyota 
announced a five-year, $lbn investment in Silicon Valley. [cci] Ford announced a JV 
with Google. [ccii] and BMW's head of R&D declared that in five years, his division 
had to transition from a department of a mechanical engineering company to a 
department of a tech company, [cciii] It remains to be seen whether Google will seek to 
be a supplier of artificial intelligence functionality to a robust and healthy automotive 
industry, or whether it will follow Tesla's example, and lead the car industry by 
competing with it. Meanwhile, there are persistent rumours that Apple wants to become 
a car company too. 

Other affected industries 

Automotive cover represents 30% of the insurance industry, so a shift to self-driving 

cars will have a major inpact on that industry. The most obvious effect should be a 
sharp reduction in pay-outs because there will be far fewer accidents. This in turn 
should mean far lower premiums: bad news for the insurance companies, good news for 
the rest of us. 

Who will take out the insurance policy? When humans drive cars we blame them for 
any accidents, so they pay for the insurance. When machines drive, does the buck stop 
with the human owner of the vehicle, the vendor of the self-driving AI system, or the 
programmer who wrote its code? If the insured parties are Google and a handful of 
massive competitors, then the negotiating position of the insurance companies will 
deteriorate sharply from the present situation where they are “negotiating” with you and 

Warren Buffet ascribes some of his enormous success as the world's best-known 
investor to his decision to avoid areas he does not understand, including industries 
based on IT. He has massive holdings in the insurance industry. Unfortunately for him, 
software is “eating the world”, [cciv] and a large chunk of the insurance industry is 
about to be engulfed in rapid technological change. Buffet acknowledges that when 
self-driving cars are established, the insurance industry will look very different, almost 
certainly with fewer and smaller players, [ccv] It is very hard to say which of today's 
players will be the winners and losers. 

The law of unintended consequences means that we cannot say how the insurance risks 
will change. Let's hope this never happens, but what if a bug - or a hacker - caused 
every vehicle in a particular city to turn left suddenly, all at the same time? How does 
an insurance company estimate the probability of such an event, and price it? Important 
issues like this - and the ethical questions we will discuss below - will slow down the 
introduction of self-driving cars. But they will not stop it. They are capable of 
resolution, just as we resolved questions about who would build the roads and who 
would have the right of way in different traffic situations in the decades after the first 
cars appeared. 

People working in insurance companies will certainly not be the only ones affected by 
the move to self-driving cars. Machines will presumably be programmed not to violate 
local parking restrictions. That will remove a significant source of income from local 
authorities: parking charges generate well over $300m a year for the city government of 
Los Angeles . [ccvi] 

Automotive repair shops will still be needed, but their business will shrink as it 
becomes restricted to maintenance and repairs necessitated by age rather than 

accidents. Happily, something similar can be said of doctors and nurses. 

Programming ethics 

Your self-driving car is travelling down the road, minding its own business, when a 
child, unpredictably, dashes across the street ahead of you. Calculating at super-human 
speed, it analyses the only three available options: maintain direction, turn right or turn 
left. Even though it has already applied the brakes far quicker than any human could 
have, it forecasts (correctly, of course) that these options will result in the death, 
respectively, of the child, of an innocent adult bystander, or of you, its passenger. 

Which option should it select? The question will have been answered in advance, even 
if only by default. 

With grim humour, some have suggested that the answer will vary by car. Perhaps a 
Rolls Royce will always choose to preserve its owner, while a Lada may accord its 
passenger less respect. 

What is happening here is the extension of human control over the world: the arrival of 
choice. Today, 27% of the victims of accidents are pedestrians and cyclists. What 
happens to them and the drivers of the cars which hit them is currently decided by the 
skill of the driver and blind chance. In future we will have the power to affect it, and 
with increased power comes increased responsibility. 

Driving jobs 

Clearly, self-driving vehicles will have a huge impact on society, sometimes in 
surprising ways. What impact will they have on employment? Are they indeed the 
poster child for technological unemployment? 

There are 3.5 million truck drivers in the US alone [ccvii] . 650,000 bus drivers [ccviii] 
and 230,000 taxi drivers [ccix] . How many of these jobs will be lost to machines? 

It seems inevitable that machines will drive commercial vehicles. Articulated lorries 
are driven by professional drivers whose backgrounds are checked and whose working 
hours and conditions are regulated. They cause fewer accidents per mile driven than 
cars owned by the likes of you and me. But because they are heavier, when they are 
involved in accidents they cause much more damage to life and property. It is 
inconceivable that we will continue to allow humans to do this job for which machines 
are clearly better suited. 

But driving is not the whole of the job. The people who drive trucks, delivery vans, 
buses and taxis have to deal with the myriad surprises which are thrown at them by life, 
which is an untidy business at best. If a consignment of barbed wire falls off the back of 
a truck in front, they will get out and help. They are also often responsible for loading 
and unloading their vehicles. 

Sceptics about technological unemployment could point out that planes have been flying 
by wire for decades, with human pilots in control for only around three minutes of an 
average commercial flight. We have yet to dispense with the services of human pilots. 

However, a 747 is very different to a truck travelling down Highway 66. A truck is an 
expensive vehicle, and capable of inflicting severe damage, but commercial planes are 
on a different scale: they cost many millions of dollars each, and their potential to cause 
harm was graphically and tragically demonstrated in New York in 2001. Furthermore, 
those three minutes of human control are in part due to the difficulty of resolving the 
edge cases we discussed before. In road vehicles if not in planes, we are well on the 
way towards resolving those. 

Consider the process of delivering a consignment from a warehouse to a supermarket or 
other large retail outlet. Amazon's fleet of Kiva robots show that warehouses are well 
on the way towards automation. The unloading bays at the retail end are also 
standardised for efficiency: a system which automates the entire unloading process from 
a truck into the retailer's receiving area is technically feasible today, and with the 
exponential improvement in robotics and AI, it won't be long before it is economically 
feasible as well. 

As we have seen, robots are becoming increasingly flexible, nimble and adaptable. 

They can also increasingly be remotely operated. Most of the situations a driver could 
deal with on the open road will soon be within the capabilities of a robot which does 
not need sleep, food or salary. On the rare occasion when human intervention is 
needed, the gig econom v[ccx] can probably furnish one quickly enough. 

Once it is economically feasible to replace human drivers with machines, it is a very 
short step to being economically compelling. Drivers account for 25-35% of the cost of 
a trucking operation, [ccxi] You can't escape the invisible hand of economics for long. 

In a free market, once one firm replaces its drivers the rest will have to follow suit, or 
go out of business. Of course, trade unions and sympathetic governments may try to stop 
the process in some jurisdictions. They may succeed for a while, but only by rendering 
their industry uneconomic, and burdening their customers with unnecessary costs which 
will damage them in turn. 

Other governments will take a different approach, and the competitive disadvantage 
imposed by resistance to change will become apparent. It will not be manifest only in 
the case of truck drivers, but in all areas of the economy, and regions and countries 
which do resist will find their living standards declining fast. Over time it will prove 

The science fiction writer William Gibson is reported as saying that “The future is 
already here - it's just not evenly distributed. [ccxii] ” In the Yandicoogina and 
Nammuldi mines in Pilbara, Western Australia, transport operations are now entirely 
automated, supervised from a centre in Perth, which is 1,200 miles awav. [ccxiii] 
Mining giant Rio Tinto was prompted to take this initiative by economics: the decade- 
long mining boom caused by China's enormous appetite for raw materials. Drivers 
earned large salaries in the hazardous and inhospitable environments of these remote 
mines, which made the investment case for full automation irresistible, [ccxiv] The 
economics are going in the same direction everywhere - fast. 

The automation of driving will have a major impact on the overall job market. Truck 
and delivery driver is the most common occupation in 29 US states (57% of 
them) . [coxy] It will also have the effect of alerting everyone else to the prospect of 
widespread technological unemployment. 

3.9 - Who's next? 

Low-income jobs 

The Frey and Osborne study we looked at in chapter 3.2 foresaw two waves of 
automation in the coming decade or two. “In the first wave, we find that most workers 
in transportation and logistics occupations, together with the bulk of office and 
administrative support workers, and labour in production occupations, are likely to be 
substituted by [machines] .’’ [ccxvi] It makes intuitive sense that lower income jobs 
would be less cognitively demanding, and hence easier to automate. 

Food service 

Automation is not new to the retail industry. Automated Teller Machines (ATMs) took 
over the job of dispensing cash in banks years ago, and self-service checkouts are 
familiar sights in supermarkets. Neither is it new in food service specifically. Few 
people in major cities now buy their sandwiches bespoke from a human who prepares 
their food in front of them. Far more buy their lunch in ready-made packages. 

Back in 1941, the Automat chain served half a million American customers a day, 
dispensing macaroni cheese, baked beans and creamed spinach through cubby-holes 
with glass doors, [ccxvii] The chain declined in the 1970s with the rise of fast-food 
restaurants serving better-tasting food, such as Burger King and of course McDonalds. 
But these chains themselves are now discovering the economic appeal of automation. 

Chili's Bar and Grill is rolling out a tablet ordering system, and Applebee's began 
delivering tablets to all its 1,800 restaurants in 2014. [ccxviii] There has been heated 
political debate about whether these and similar initiatives are prompted by increases in 
minimum wage levels, but the truth is that AI systems and robots are rapidly becoming 
more efficient at cost levels which humans can never hope to match. 

Customer preference 

Cost saving is not the only reason for this kind of automation. In many situations, 
humans prefer to transact with machines rather than other humans. It can be less time- 
consuming, and require less effort. It can also make a service available for longer 
hours, perhaps 24 hours a day. Bank ATMs are the classic example. Another are the 

automated passport control systems now installed at many airports, which many people 
opt to use in preference to the manned channels. 

A report published in April 2015 by Forrester, a technology and market research 
company, claimed that 75% of procurement professionals and other people buying on 
behalf of businesses (i.e., B2B buyers) prefer to use e-commerce and buy online rather 
than deal with a human sales representative. Once the buyers have decided what they 
want, the percentage rises to 93%. [ccxix] Forrester pointed out that many vendors were 
ignoring this fact and obliging customers to speak to a human. This is no doubt partly at 
least because human sales people are currently much better able to up-sell the buyer, but 
this is also one of the reasons why buyers prefer e-commerce. Forrester argued that 
companies which wait too long to offer good e-commerce channels risk losing market 
share to more digitally-minded competitors. 

Call centres 

We are still at the very early stages of introducing artificial intelligence to call centres. 
For many of us, dealing with call centres is one of the least agreeable aspects of modern 
life. It normally involves a good deal of waiting around, listening to uninspiring hold 
music, followed by some profoundly unintelligent automated routing, and finally a 
conversation with a bored person the other side of the world who is reading from a 
script written by a sadist. 

One of the leaders in introducing genuine AI to call centres is Swedbank, one of 
Sweden's biggest banks, with 9.5m customers and 160,000 employees. It has 700 
people working in contact centres, which handle 2m customer calls each year. It has 
worked with the American software company Nuance to introduce a basic AI called 
Nina. [ccxx] which learns what customers want and how best to help them by 
assimilating searches made on the company website and enquiries made at the contact 
centres. [ccxxi] In December 2015, Nina was handling 30,000 calls a month, and taking 
care of many of the straightforward transactional calls - like transferring money from 
one account to another - which were previously clogging up the call centres. The aim is 
to free up the agents in the contact centres to concentrate on more complicated activities, 
like taking out a mortgage. But even taking out a mortgage isn't rocket science. Given 
exponential progress, if Nina can handle transfers today, it will surely be able to handle 
mortgage applications before long. 

Manual work 

Occupations requiring physical labour will take longer to automate than clerical and 
administrative jobs because getting robots to be dextrous and flexible is surprisingly 
hard. As we saw in chapter 3.7, progress is rapid, but much remains to be done. 

Manual work in routine, repetitive environments like assembly lines will continue to 
see machines taking over, but physical labour in unstructured environments like building 
sites will remain the preserve of human workers for a while longer. 

Manufacturing accounts for over a third of China’s GDP, and employs more than 100 
million of its citizens. Historically, China’s competitive strength in manufacturing has 
been its low wage costs, but this is changing fast: wages have grown at 12% a year on 
average since 2001, and Chinese manufacturers are embracing automation 
enthusiastically. As we saw in chapter 2.3, China is now the world’s largest market for 
industrial robots, but it has a long way to go before it catches up with the installed base 
in more developed countries. 

Industrial robots are far from perfect, and manufacturers have under-estimated the 
progress still required. In2011 the CEO of Foxconn, a $130bn-turnover Taiwanese 
manufacturer that is famous for making iPhones, declared a target of installing a million 
robots by 2014. The robots failed to perform as he hoped, and the actual installation 
rate has been much slower. But the robots are improving fast. [ccxxii] 

The professions 

It is certainly not only low-paid, relatively low-prestige service jobs that will be 
automated. The professions are vulnerable too: lawyers, doctors, architects and 
journalists. Sometimes accused of being conspiracies against lay people, these are 
protected occupations, with demanding entry requirements and restrictions on the 
number of trainees who can join the professions each year. They have commanded 
prestige and high salaries, but that may be about to change. 


Nuance, the company behind Swedbank's Nina call centre AI, offers services for 
journalists, helping them create interviews and articles faster. But Narrative Science, a 
company established in Chicago in 2010, has an AI system which writes articles 
without human help. Called Quill, it already produces thousands of articles every day 
on finance and sports for outlets like Forbes and Associated Press fAPT [ccxxiii] Most 
readers cannot identify which articles are written by Quill and which by human 
journalists, and Quill is much faster. 

Quill starts with data - graphs, tables and spreadsheets. It analyses these to extract 
particular facts which could form the basis of a narrative. It then generates a particular 
plan, or narrative for the article, and finally it crafts sentences using natural language 
generation software. 

A British company called Arria offers the same functionality, but sells mainly to 
corporations trying to make sense of the tsunami of data which threatens to overwhelm 
them. [ccxxiv] 

In the short term, Quill has not rendered thousands of journalists redundant. Instead it 
has sharply increased the number of niche articles being written. Newspaper revenues 
have declined sharply since the turn of the century, as classified ads for jobs, houses 
and cars migrated online. News services like AP increased the daily quota of articles 
for each journalist, cut back the number of journalists they employed, and reduced the 
number of articles they produced on, for instance, the quarterly earnings reports of 
particular companies. Quill and similar services have enabled them to reverse that 
decline. AP now produces articles on the quarterly reports of medium-sized companies 
that it gave up covering in such detail years ago. 

Kristian Hammond, founder of Narrative Science, forecast in 2014 that in a decade, 

90% of all newspaper articles would be written by AIs. However, he argued that the 
number of journalists would remain stable, while the volume of articles increased 
sharply. Eventually, articles could become tailored for particular audiences, and 
ultimately for each of us individually. For instance, an announcement by a research 
organisation that inflating your car tyre correctly could reduce your spend on petrol by 
7% could be tailored - perhaps with the help of your Digital Personal Assistant - to 
take into account your particular car, the number of miles you drive each week, and 
even your style of driving. (Although of course by then you will perhaps not do much of 
the driving yourself anyway.) 

The prediction that the number of human journalists will remain stable sounds 
reassuring, and indeed you would expect someone marketing an automating technology 
to say that. Given the exponential improvement of AIs, it is a brave prediction. 

TV presenters should also be feeling nervous. In December 2015, Shanghai's Dragon 
TV featured Xiaoice, (pronounced Shao-ice) an AI weather presenter with a remarkably 
life-like voice, [ccxxv] based on the Mandarin version of its Cortana digital assistant 
software. Audience feedback was positive. [ccxxvi] 

Other writers 

Not everyone who spends their working days crafting crisp sentences is a journalist. 
They might be PR professionals, or online marketers, for instance. A company called 
Persado claims that marketing emails drafted by its AI have a 75% better response rate 
than emails written by human copywriters, [ccxxvii] Citibank and American Express are 
customers as well as investors. 

In January 2016 a researcher at the University of Massachusetts announced an AI which 
can write convincing political speeches for either of the two main US political parties. 
The system learned its craft by ingesting and analysing 50,000 sentences from 
Congressional debates. [ccxxviii] 

Two other professions have come under particular scrutiny with respect to their 
susceptibility to machine automation: the law and healthcare. Let's look at these in turn. 


Whatever Hollywood thinks, most lawyers do not spend their days pitting their razor- 
sharp wits against equally talented adversaries in front of magisterial judges, eliciting 
gasps of admiration from around the courtroom as they produce the winning argument 
with a flourish. Most of the time they are reading through piles of very dry material, 
looking for the thread of evidence which will convict a fraudster, or the poorly drafted 
phrase which could undermine the purpose of a contract. 


Many lawyers get a lot of their on-the-job training through the “discovery” process. 
Known as “disclosure” in the UK, this is a pre-trial process in civil law in which both 
sides must make available all documents which may affect the outcome of the case. An 
analogous process takes place in the “due diligence” phase of a corporate merger or 
acquisition (M&A), in which teams of junior lawyers (and accountants) spend weeks 
locked away in data rooms, reading through material which can run into millions of 
documents, looking for something which would clinch the case, or, in the case of M&A 
work, provide a reason to terminate or renegotiate the deal. 

Looking for a needle of fact in a haystack of paper is work more suited to a machine 
than a man. And although lawyering is a very conservative profession, there are signs 
that it understands what is coming better than some others. RAVN Systems is the British 
AI company behind an AI system called Ace, which reads and analyses large sets of 

unstructured, un-sorted data. It produces summaries of the data, and highlights the 
documents and passages of most interest according to the pre-set criteria, [ccxxix] When 
one of the UK's largest law firms started working with Ace, it was regarded as 
pioneering, experimental, and somewhat risky. Two years later that law firm was 
promoting its own services to potential new clients on the basis that it knew best how to 
exploit the advantages of RAVN Ace. Bear in mind that two years is a very short time 
for anything at all to happen in the legal industry! 

Typically, a new client's data will present a new set of challenges. It usually takes a 
few days to train the system how to read the data, which currently involves human 
intervention. Once the training is complete, the work proceeds without human 
involvement, and the system will finish the work much faster than human lawyers could. 
This means that law firms are having to work out new ways of billing their clients: the 
old system of hourly rates is under challenge. 

Revealing the iceberg 

Forward-thinking lawyers are actually excited about the arrival of this sort of 
automation. Rather than fearing that it will destroy the jobs of junior lawyers, making it 
impossible for young people to learn the profession, they believe it will increase the 
amount of cases that can be handled. To illustrate this, imagine a large supermarket 
chain that wants to know the implication of making a small change to the employment 
contracts of all its in-store employees - tens of thousands of people. Previously, its 
employment law firm would have said that this task could not be undertaken cost- 
effectively with any degree of rigour. RAVN Ace and systems like it make this kind of 
work possible, opening up whole new avenues of work for law firms. It is like standing 
nervously on a body of ice, thinking you are only separated from freezing water by a thin 
layer, and suddenly discovering that in fact you are standing on an iceberg, with a huge 
mass of previously unknown solid ice beneath your feet. 

As Greg Wildisen, MD of Neota Logic, a firm providing an AI platform for lawyers, 
puts it, “So many legal questions go 'un-lawyered' today that there is enormous scope to 
better align legal resources through technology rather than fear losing jobs.” [ccxxx] 

So in the short and medium term, machine automation of white-collar jobs opens up vast 
new areas of work that can be undertaken, and doesn't throw the incumbent humans out 
of work. They are still needed to train the system at the start of a large new assignment, 
and to process more complicated documents. 

But as RAVN Ace and its successors improve - at an exponential rate, of course - they 

will be able to take on more and more of the sophisticated and demanding aspects of the 
lawyers' work. No-one can be absolutely sure yet whether this process will hit a wall 
at some point, leaving plenty of work for humans, or whether it will continue to the 
point where there are very few jobs left for humans. My own view is that within a few 
short decades, the machines are coming for most of our jobs. 

The short-term explosion of work which happens as the iceberg of latent demand is 
revealed can give us a false sense of security. The phenomenon of automation leading 
to job creation is sometimes called the automation paradox. [ccxxxi] But the paradox 
may turn out to be short-lived. 


Another fairly basic form of legal work is the completion of boilerplate (standard) 
forms to establish companies, initiate a divorce, register a trademark, request a patent 
and so on. A company called LegalZoom was established in 2001 to provide these 
services online, and increasingly, to automate them. LegalZoom now claims to be the 
best-known law brand in the US. [ccxxxii] and in 2014 the private equity firm Permira 
paid $200mto become its largest shareholder. Another company, Fair Document, helps 
clients complete forms for less than $1,000, one-fifth the amount it would have 
previously cost. [ccxxxiii] 

More sophisticated work 

At the other end of the spectrum from the “grunt” work of discovery and filling out legal 
forms, one of the most sophisticated and important jobs that senior and successful 
lawyers are asked to undertake is to estimate the likelihood of a case winning. The 
advice is vital as it will determine whether large amounts of money are spent. A team 
led by Daniel Martin Katz, a law professor at Michigan State University, developed an 
AI system that analysed 7,700 US Supreme Court cases. It predicted the verdicts 
correctly 71% of the time. [ccxxxiv] 

Another job for experienced lawyers in common law jurisdictions such as the US and 
the UK is identifying which precedent cases to deploy in support of litigation. A system 
called Judicata uses machine learning to find the relevant cases using purely statistical 
methods, with no human intervention, [ccxxxv] 

Will entire sections of the legal industry be automated in the next few years or 
decades? How about patent lawyers, for instance? Senior patent lawyers are highly 

skilled and articulate people, but much of the work involved in securing a patent is 
routine and could perhaps be automated. In November 2015 I took part in a debate at 
the UMAX cinema in London’s Science Museum. The motion was “This House believes 
that within 25 years, a patent will be applied for and granted without human 
intervention.” Patent lawyers comprised a good part of the audience, and although the 
motion was vigorously opposed by two senior patent lawyers, the motion was passed. 
Not exactly turkeys voting for Christmas, but certainly food for thought. 


Doctors are a scarce resource. Only bright and dedicated people are admitted to the 
relevant university and post-graduate courses, and these courses demand many years of 
hard study. Hospitals and local surgeries are organised to maximise the availability of 
this resource, but some critics argue that they are organised for the benefit of the doctors 
rather than the patients. In 2015, senior doctor and medical researcher Eric Topol 
published a book called “The Patient Will See You Now”, which he argues should 
become the mantra for the profession, replacing the current one, which he says is “the 
doctor will see you now”. 

Suggesting acerbically that the initials MD stand for Medical Deity, Topol accuses 
many doctors of being arrogant and paternalistic towards their patients, assuming they 
are unable to understand the detailed information regarding diagnoses, and withholding 
information from patients so as not to upset them. He believes that the digital revolution 
will start to overturn this unsatisfactory state of affairs, as it will place cheap and 
effective diagnostic tools in the hands of patients. 

Better and cheaper diagnostics 

In April 2016, researchers at Indiana University announced that a test of open-source 
machine learning algorithms on 7,000 free-text pathology reports from 30 hospitals 
yielded equal or better diagnoses than humans had made. The computers were also 
faster and cheaper, [ccxxxvi] 

This sort of technology will become much more widely available. A British startup 
called Babylon charges customers £5 a month for phone (and videophone) access to a 
dedicated team of doctors. Before a doctor comes on the line, the patient is triaged by a 
machine, [ccxxxvii] As AI improves, the role of the human doctor in this process will 
continuously be reduced. 

Smartphones are increasingly able to gather medical data about us, and perform basic 
analysis. By attaching cheap adapters to their phones, patients can quickly take their 
blood pressure, sample their blood glucose, and even perform an electrocardiagram. 
Your breath can be sampled and digitised, and used to detect cancer, or potential heart 
problems. Your camera's phone can help screen for skin cancers. Its microphone can 
record your voice, and that data can help gauge your mood, or diagnose Parkinson's 
disease or schizophrenia. 

All this data can be analysed to a certain level within the phone itself, and in many 
cases that will suffice to provide an effective diagnosis. If symptoms persist, or if the 
diagnosis is unclear or unconvincing, the data can be uploaded into the cloud, i.e., to 
server farms run by companies like Amazon and Google. The heart of diagnosis is 
pattern recognition. When sophisticated algorithms compare and contrast a set of 
symptoms with data from millions or even billions of other patients, the quality of 
diagnosis can surpass what any single human doctor could offer. 

Ross Crawford and Jonathan Roberts are professors of orthopaedic research and 
robotics respectively at Queensland University of Technology. In an article in January 
201 6. [ccxxxviii] they argued that doctors need to understand that diagnostic services 
can be made available more cheaply with the assistance of machine intelligence, and 
reach all the patients who need them, not just those in rich countries who are already 
manifesting symptoms. 

They don't think this will render doctors unemployed. As with the law, there is an 
iceberg of unmet healthcare needs - needs which automation and machine intelligence 
can satisfy. Formed in Mumbai in 1996, Thyrocare Technologies is the world's largest 
thyroid testing laboratory. Its founder, Dr A Velumani, had the insight that 90% of 
people who could benefit from diagnostic tests were not receiving them because they 
were too expensive, so the tests were restricted to those already manifesting symptoms 
of disease. He established Thyrocare to address this latent demand, and now it 
processes 40,000 samples a dav. [ccxxxix] 

As this iceberg is revealed, the healthcare industry - like the legal industry - can 
perform a far better job by reaching many more people at greatly reduced average cost. 
At first, there will be just as much need for doctors as before: they will continue to 
carry out the more sophisticated diagnoses, while machines (possibly deployed by less 
highly-trained people) deliver the routine work. But as we keep observing, the 
machines are getting smarter at an exponential rate. In time, what is to stop them 
performing the doctors' other roles as well, and doing them better, faster and cheaper? 


If machines can diagnose, can they go on to prescribe, and to fulfil prescriptions? The 
University of California in San Francisco has installed a robot pharmacist which is 
reported to have prepared 6,000,000 prescriptions with only one error - a track record 
which is 60,000 times better than human pharmacists, [ccxl] 

Keeping current 

As we saw in chapter 3.4, machines are reaching parity with humans in pattern 
recognition, and will quickly become much better. They are already much better than 
human professionals at keeping up with new developments in their field. 

A human doctor would have to read for 160 hours a week just to keep up with the 
published medical research. This is clearly impossible for a human, but machines have 
no such bandwidth restriction. IBM is pushing aggressively into the medical industry 
with its Watson AI system. According to Samuel Nessbaum of Wellpoint, a private 
healthcare company, Watson’s diagnostic accuracy rate for lung cancer is 90%, which 
compares favourably with 50% for human physicians. [ccxli] 

IBM has come in for criticism for pretending that Watson is a unitary system rather than 
a kludge of different systems which can be mixed and matched according to need. It is 
also accused of scaling back its ambition by tackling much smaller projects than the 
“moonshots” it was originally earmarked for, like curing cancer, [ccxlii] Kris 
Hammond, the founder of Narrative Science whom we met when discussing journalists, 
says that “everybody thought [winning Jeopardy] was ridiculously inpossible, [but 
now] it feels like they're putting a lot of things under the Watson brand name - but it isn't 
Watson. ’’ [ccxliii] In March 2016, DeepMind founder Demis Hassabis went as far as to 
say that Watson is essentially an expert system as opposed to deep learning one. [ccxliv] 

IBM is unfazed by this kind of criticism. It says that Watson is now being used by 
hundreds of companies to solve particular problems - companies like the Australian 
energy group Woodside, which used it to review 20,000 documents from 30 years of 
engineering projects to identify, for instance, the maximum pressure that a certain type 
of pipeline can withstand. It might be a form of marketing sleight of hand to apply the 
Watson brand to all these applications, but the company spent a great deal of time and 
money to create that brand, and it would be unreasonable to expect it not to try and 
recoup that investment. 

That said, IBM is developing a new brand for its commercial AI offering. Celia stands 
for Cognitive Environments Laboratory Intelligent Assistant, and it seems to be a more 
user-friendly front end, enabling business analysts, for instance, to interact with it by 
speech, and by manipulating virtual objects in an augmented reality field. [ccxlv] 

And IBM is still pursuing moonshots, in the medical field and elsewhere. As we have 
noted several times, machine learning is fuelled by data. In October 2015, IBM paid 
$lbn for Merge Healthcare, a company with 30 billion medical images. [ccxlvi] and 
$2bn for the digital assets of The Weather Company, to build a weather forecasting 
service. At the end of the year it unveiled Avicenna, a product of the Watson healthcare 
business unit designed to help radiologists prioritise which images to review, and help 
them make diagnoses . [ccxlvii] The interesting question is, how long before at least 
some of those radiologists turn out to be superfluous to the process. 


You might think that the hands-on physical and frankly messy business of surgical 
operation will be undertaken by humans rather than machines for the foreseeable future. 
Probably not. One of the most highly skilled professionals in the emergency suite is the 
anaesthetist, and Johnson & Johnson has an automated version called Sedasys which, 
despite fierce opposition from the profession, has FDA approval to provide the 
anaesthesia in less challenging procedures like colonoscopies. It has carried out 
thousands of operations in Canada and the USA. [ccxlviii] In March 2016, Johnson & 
Johnson announced that it was exiting the Sedasys business due to sluggish sales, 
despite the machine costing $150 per operation whereas a human anaesthetist costs 
$2.000. [ccxlix] This will certainly not be the last setback in the progress of the 
machines, but in the long run the economic facts will prevail - although perhaps more 
slowly in industries where the normal rules of the market do not always apply. 

In May 2016 an academic paper announced that a robotic surgeon had out-performed 
human peers. The Smart Tissue Autonomous Robot (STAR) operated on pig tissue and 
did the job better, although four times more slowly, than humans operating alone - and 
also better than humans aided by the semi-robotic Da Vinci system, [cel] 


Teachers are the active ingredient in education - at school level, anyway. Studies have 
shown repeatedly that the quality of teaching makes an enormous difference to how well 
a student performs at school and afterwards. But schools cannot afford enough of them, 

governments burden them with bureaucracy, and most countries’ cultures under-value 

What happens when the learning of every pupil is monitored minutely by artificial 
intelligence? When every question she asks and every sentence she writes is tracked 
and analysed, and appropriate feedback is provided instantly? Teachers will play the 
role of coach instead of instructor, but as with the other professions, their scope for 
contribution will shrink. 

The beachhead for AI in education is marking, also known as grading. This is the bane 
of many teachers’ lives, and they will welcome an assistant which can relieve them of 
the duty. A company called Gradescope marks the work of 55,000 students in 100 US 
universities, marking simple, multiple-choice types of test. It raised $2.6 million in 
April 2016 to develop its product into complex questions and essavs. [ccli] Large 
corporates which provide education services like Pearson and Elsevier are moving in 
the same direction. 

Towards the end of 2015, 300 students at Georgia Institute of Technology were, 
unbeknownst to them, guinea pigs in an experiment to see whether they would notice that 
one of their nine teaching assistants was a robot. Only ever in contact via email, they 
would ask questions like “Can I revise my submission to the last assignment?” and 
receive answers back like “Unfortunately there is not a way to edit submitted 
feedback.” None of the students noticed that Jill Watson, named after the IBM Watson 
system “she” ran on, was in fact an AI. [cclii] 

Financial services 

The finance sector is an obvious target for machine intelligence, with high-value (and 
high-priced!) services based on vast amounts of data. Human equity analysts and 
brokers will increasingly struggle to provide value in the face of competition with 
machines which can ingest all the relevant data, and never forget any of it. The 
provision of advice to investors is also migrating to machines, with systems like SigFig 
incorporating a client's risk appetite and investment style into its algorithms' analysis of 
low-cost opportunities and recommendations. [ccliii] Similar so-called “robo-adviser” 
services are available from Betterment, Wealthfront and Vanguard, [ccliv] 

These services deploy primitive forms of AI at the moment. According to market 
research firm Preqin . thousands of hedge funds, managing $200bn of assets, use 
computer models in most of their trades. But they are using traditional statistical 
methods rather than AI which learns and evolves. This is changing. Bridgewater, the 

world's largest hedge fund, hired David Ferrucci away from IBM, where he had project 
managed the development of the version of Watson which beat Ken Jennings at 
Jeopardy, [cclv] January 2016 saw the inaugural trades of Aidyia, a hedge fund based in 
Hong Kong whose chief scientist is Ben Goertzel, one of the leading researchers in 
artificial general intelligence. [cclvi] 

In May 2016, the AHL hedge fund announced that it was stepping up its use of machine 
learning, having researched it for five years and deployed it experimentally for three. 
With $ 19 billion under management, AHL is the largest part of the MAN Group, which 
in turn is the world’s largest publicly-traded hedge fund. [cclvii] 

“The human mind has not become any better than it was 100 years ago, and it’s very 
hard for someone using traditional methods to juggle all the information of the global 
economy in their head,” says David Siegel of Two Sigma, another hedge fund which 
uses AI. “The time will come that no human investment manager will be able to beat the 
computer.” [cclviii] 

“Algo trading” has many critics in financial circles, who point out that they chase 
spurious correlations (such as the fact that divorce proceedings in Maine have 
consistently tracked sales of margarine), and that they can move markets in ways that are 
impossible to follow and are potentially dangerous. But in a financial world become so 
complex that mere humans can no longer follow it, they may not only be inevitable, but 
also necessary. David Siegel says “People talk about how robots will destroy the 
world, but I think robots will save it.” 

What is known as “fintech” is one of the hottest areas for VC investment at the time of 
writing, and banks are spending considerable amounts of time and energy on working 
out where the most powerful disruption to their business models will come from, and 
whether they can do the disrupting themselves rather than be its victims. Banking, 
especially retail banking, has traditionally been a conservative, slow-moving industry, 
but the pace is picking up. Goldman Sachs reports that 40% of all US cheques are now 
processed electronically, despite that service being only four years old. jcclix] 

Top analysts are among the highest-paid people in investment banks. They are targeted 
by a number of fintech companies like Kensho, which sorts through thousands of data 
sets to produce reports in minutes which would take skilled humans days. For instance, 
asking the system about Syrian Civil War will generate a report showing its impact on 
companies, currencies and commodities in as many countries as you like. Kensho’s 
founder, Daniel Nadler, thinks between a third and a half of finance employees will be 
redundant within a decade. [cclx] 

One of the drivers of AI use by financial services firms is the ever-growing and 
increasingly complex web of compliance requirements imposed by governments and 
regulators. Systems like IPSoff s Amelia help insurance firms and other financial 
services companies to navigate this web and make sure the forms and procedures used 
by staff are up-to-date, [cclxi] 

Global banks are regularly fined hundreds of millions of dollars for carrying out illegal 
or sanctioned trades. Standard Chartered’s regulatory costs rose 44% in 2015 to 
$447m as it was obliged to hire thousands of additional staff to deal with compliance 
requirements. In March 2016 it announced a major investment in AI systems to oversee 
its traders’ behaviour, and to match their activities against regulatory norms . [cclxii] 

It seems that managers in financial services are becoming aware of the potential threat 
to their livelihood. A survey of 1,700 managers in 17 different industries carried out in 
the autumn of 2015 by the consulting firm Accenture revealed substantial anxiety about 
automation. Overall, a third of the managers feared that intelligent machines threatened 
their jobs, with the level rising to 39% among senior managers. Unsurprisingly, the 
anxiety was highest (50%) among managers in the technology sector, but it was also 
high in banking, at 49%. [cclxiii] 

3.10- Jobs or no jobs 

The question 

In chapter 2 we saw that previous rounds of automation in the industrial revolution did 
not cause unemployment in the long-term - although the long term is fairly long in this 
context: the Engels pause that we came across in chapter 2.4 lasted at least a generation 
(a quarter-century). 

In chapters 3.1 to 3.3 we reviewed in brief the arguments of those who believe it is 
different this time - and those who believe the opposite. 

In chapter 3.4 we considered the state of the art of artificial intelligence, and in chapter 
3.5 we saw how dramatic the impact of exponential growth can be. (No apologies for 
repeating that point: it is critical.) Then in chapter 3.7 we reviewed the likely evolution 
of technologies which are associated with AI, and caused or enabled by it. 

Finally, in chapters 3.8 and 3.9 we discussed how various occupations could be 
automated, starting with the poster child of driving vehicles, and concluding with the 
privileged elites in the professions. 

Now we are ready to address head-on the first great question posed by machine 
intelligence automation, which is this: Is it different this time? Will the automation of 
jobs by machine intelligence lead to widespread, lasting unemployment? 

For the answer to be negative, we will have to dramatically increase the supply of jobs 
which are for some reason immune to automation by machine intelligence. 

Jobs, not work 

It is important at this point to distinguish between jobs and work. Physicists define 
work as the expenditure of energy to move an object. [cclxiv] but what we mean by it 
here is the application of energy in pursuit of a project. That energy could be physical, 
mental, or both. Work could be instigated by an employer, but it also could be purely 
personal: building or decorating a home, pursuing a hobby, or an unpaid community 

A job, on the other hand, is always paid labour for the purpose of this discussion. It 
might be a salaried occupation with a single, stable employer, or it could involve self- 

employment, or freelance activity. Your job is the way you participate in the economy, 
and earn the money to buy the goods and services that you need to survive, and enjoy a 
good standard of living. 

If a machine carries out a job, there is no point a human replicating the work it is doing: 
she will not be paid, so she will have to look for some other way to generate an 

The gig economy 

We saw in chapter 3.2 how consultants at McKinsey noted that jobs can be analysed 
into tasks, some of which can be automated with current machine intelligence 
technology, and some of which cannot. This is an important insight and suggests that 
jobs will be sliced and diced, with some tasks being automated, and other tasks being 
retained by the human who previously did the whole job. 

Some would argue that this process is already under way. Parts of the economies of 
developed countries are being fragmented, or Balkanised, with more and more people 
working freelance, carrying out individual tasks which are allocated to them by 
platforms and apps like Uber and TaskRabbit. 

There are many words for this phenomenon: the gig economy, the networked economy, 
the sharing economy, the on-demand economy, the peer-to-peer economy, the platform 
economy, and the bottom-up economy. 

Is this a way to escape the automation of jobs by machine intelligence? To break jobs 
down into as many component tasks as possible, and preserve for humans those tasks 
which they can do better than machines? Probably not, for at least two reasons. First, it 
is precarious, and secondly, the machines will eventually come for all the tasks. 

Working for yourself can seem an appealing prospect if your current job is a poorly- 
paid round of repetitive and boring activities. There is freedom in choosing your own 
hours of work, and fitting them around essential parts of your life like children and 
hangovers. There is freedom in choosing who you work with, and in not being subject 
to the arbitrary dictates of a vicious or incompetent boss, or the unfathomable rules and 
regulations of a Byzantine bureaucracy. 

If you are lucky enough to be exceptionally talented, or skilled at a task which is in high 
demand, then you really can choose how and when you work. But freelancing can have 
its downsides too. Many freelancers find they have simply traded an unreasonable boss 

for unreasonable clients, and feel unable to turn down any work for fear that it will be 
the last commission they ever get. Many freelancers find that in hindsight, the 
reassurance of a steady income goes a long way to compensate for the 9 to 5 routine of 
the salaried employee. 

Whether or not the new forms of freelancing opened up by Uber, Lyft, TaskRabbit, 
Handy and so on are precarious is a matter of debate, especially in their birthplace, San 
Francisco. Are the people hired out by these organisations “micro-entrepreneurs” or 
“instaserfs” - members of a new “precariaf’, forced to compete against each other on 
price for low-end work with no benefits? Are they operating in a network economy or 
an exploitation economy? Is the sharing economy actually a selfish economy? 
Whichever side of this debate you come down on, the gig economy is a significant 
development: a survey by accounting firm PricewaterhouseCoopers found that as many 
as 7% of US adults were involved in it. [cclxv] 

But our concern here is not whether the gig economy is a fair one. It is whether it can 
prevent the automation of jobs by machine intelligence leading to widespread 
unemployment. The answer to that is surely No: as time goes by, however finely we 
slice and dice jobs into tasks, more and more of those tasks are vulnerable to 
automation by machine intelligence as it improves its capabilities at an exponential rate. 

What tasks, if any, will machines remain unable to automate for the foreseeable future? 


A computer first beat the best human at chess back in 1997. Deep Blue was one of the 
most powerful computers in the world when it beat Gary Kasparov; the match was close 
and the result was controversial. Today, a programme running on a laptop could beat 
any human. 

But Kasparov claims that a very good human chess player teamed up with a powerful 
chess computer can beat a second chess computer playing on its own. Humans can 
undermine the game of a computer by throwing in some surprise moves which don't 
make much sense in the short term, or by deploying an intuitive strategy. Matches 
between humans working with computers are called advanced chess, or centaur chess. 
Kasparov himself initiated the first high-level centaur chess competition in Leon, in 
Spain, in 1998, and competitions have been held there regularly ever since. Tyler 
Cowen (one of the sceptics about machine automation that we met in chapter 3.3) 
explores this form of chess extensively in his book, “Average is Over”. 

Some people believe this phenomenon of humans teaming up with computers to form 
centaurs is a metaphor for how we can avoid most jobs being automated by machine 
intelligence. The computer will take care of those aspects of the job (or task) which are 
routine, logical and dull, and the human will be freed up to deploy her intuition and 
creativity. Engineers didn’t become redundant just because computers replaced slide 
rules. Kevin Kelly, founder of Wired magazine, puts it more lyrically: machines are for 
answers; humans are for questions . [cclxvi] 

The trouble is that the intuition and creativity which we humans bring to tasks and jobs 
is largely a matter of pattern recognition, and machines are getting better at this at an 
exponential rate. A doctor may be happy to delegate the routine diagnosis of a cold or a 
flu to a machine which can do it better than she can, if she gets to retain the more 
interesting and challenging diagnostic work. But what is to stop the machine overtaking 
the doctor in the more difficult cases as well? The lawyer is in the same boat: the 
tedious business of sifting through a haystack of documents looking for the needle of 
evidence is already being outsourced to machines. The more interesting and demanding 
task of devising a legal strategy is likely to follow suit. 

Admittedly, there may need to be some level of human supervision of machine work 
until the machines acquire a degree of common sense. Before then, the blindly logical 
thought processes of a machine will not realise when a data glitch or a software bug has 
generated a bizarre conclusion which is unworkable or dangerous. But as we saw in 
chapter 3.5, the founding father of deep learning thinks that machines with common 
sense will appear in a decade or so. This does not mean that they will acquire 
consciousness, but merely that they will create internal models of the external world 
which will enable them to appreciate the impacts of glitches and bugs, just as we do. 

Machines have already made considerable progress automating routine tasks, and 
indeed whole occupations where all the tasks are routine. As their performance 
improves they will increasingly take over tasks and jobs which are non-routine. 

In response to a survey published in May 2016, the veteran AI researcher Nil Nilsson 
suggested laconically that before long, machines would be singing the song Irving Berlin 
composed for the 1946 Broadway musical “ Anni e Get Your Gun”. The lyric is 
“Anything you can do, I can do better. I cam do anything better than you.” [cclxvii] 

The human touch 

Some observers think that our salvation from machine intelligence automation lies in our 

very humanity. Our social skills, and our ability to empathise and to care mean that we 
carry out tasks in a different way than machines do. Machines are by definition 
impersonal, the argument goes, and this renders them unsuitable for some types of job. 

David Deming, a research fellow at the US National Bureau of Economic Research, 
believes we are already seeing the implications of this. In a report published in 2015 
he claimed that the fastest growth in US employment since as long ago as 1980 has been 
in jobs requiring good social skills. Jobs requiring strong analytical abilities but no 
social skills have been in decline - with the implication that they are already being 
automated, [cclxviii] 

Unfortunately, it isn't true that humans want to deal with other humans whenever 
possible. The first automatic deposit machine, the Banlcograph, was installed in a bank 
in New York in 1960, but it was rejected by its intended customers. Its inventor, Luther 
Simjian, explained that “The only people using the machines were prostitutes and 
gamblers who didn’t want to deal with tellers face to face,” and there were not enough 
of them to make the machines a worthwhile investment, [cclxix] The first cash 
dispensing machine, or ATM, was installed in a bank in North London in June 1967. At 
first, people were again hesitant to use it, but that changed when they realised they no 
longer had to queue for their cash, and they could access it when the banks were closed 
(which was most of the time, in those days). Very quickly, people showed a marked 
preference for the machine over the human bank teller, [cclxx] 

Nursing is an occupation long associated with caring people. Images of Llorence 
Nightingale emoting as she nursed the wounded of the Crimean War are deeply 
ingrained in the profession's self-image. But there is evidence from Japan, Denmark 
and elsewhere that robots make perfectly acceptable companions for sick people, and 
are sometimes preferable to their human equivalents. The Paro is a robotic seal 
developed for use in hospitals. Cute-looldng, with big black eyes and covered in soft 
fur, it contains two 32-bit processors, three microphones, 12 tactile sensors, and it is 
animated by a system of silent motors. It recharges by sucking on a fake baby pacifier. 

The Paro cost $ 15m to develop; it distinguishes between individual humans, and repeats 
behaviours which appear to please them. [cclxxi] It has proved especially popular with 
patients suffering from dementia. As Shannon Vallor, a philosophy professor at Santa 
Clara University remarked, “People have demonstrated a remarkable ability to transfer 
their psychological expectations of other people’s thoughts, emotions, and feelings to 
robots.” [cclxxii] 

So humans are happy to interact with machines more often that we might intuitively 

expect. Furthermore, machines are much better at understanding humans than we might 

Robot therapist 

The US Army has a big problem with post-traumatic stress disorder (PTSD) among 
veterans, not least because soldiers don't like to admit they have it. DARPA funds 
research at the University of Southern California to develop online therapy services, 
and the latest result is an online virtual therapist called Ellie. [cclxxiii] She is proving 
to be better than human therapists at diagnosing PTSD. 

There are two reasons for this. First, soldiers feel less embarrassed discussing their 
feelings with an entity they know will not judge them. In one test, 100 subjects were 
told that Ellie was controlled by a human, and another 100 were told that it was a 
robot. This second group displayed their feelings more openly, both verbally and in 
their expressions . [cclxxiv] 

Secondly, and perhaps more interesting, Ellie gleans most of its information about what 
is going on inside the soldier's head from his facial expressions rather than from what he 
says. When talking to a human therapist, the soldier may successfully “sell” the idea 
that there is nothing wrong, because the human therapist listens closely to what he says, 
and may miss the subtle facial signals that contradict him. Counter-intuitively, people 
with depression smile just as frequently as happy people, but their smiles are shorter 
and more forced. Ellie is superb at catching this. [cclxxv] 

Most people would probably agree with David Deming when he says that "Reading the 
minds of others and reacting is [a skill that] has evolved in humans over thousands of 
years. Human interaction in the workplace involves team production, with workers 
playing off of each other’s strengths and adapting flexibly to changing circumstances. 
Such non-routine interaction is at the heart of the human advantage over machines." But 
we may soon have to re-think that. 

It is far from clear that there could ever be enough jobs in the so-called caring 
professions to employ all the people who would in previous generations have been 
drivers, doctors, lawyers, management consultants and so on. Especially if machines 
are muscling into the caring professions too. 

Made by hand 

Another way that people have suggested the human touch could preserve employment is 
that we will place a higher value on items manufactured by humans than on items 
manufactured by machines. It is hard to see much evidence of this in today's world 
outside some niche areas like hand-made cakes. [cclxxvi] Not many people today buy 
handmade radios or handmade cars. 

There are four reasons why people might prefer products and services made by humans 
rather than machines: quality, loyalty and variation and status. 

If humans produce a better product or provide a better service than machines then other 
humans will buy from them. But the argument of this chapter is that certainly in many 
areas, and probably in most, machines will produce goods and services cheaper, better 
and faster. 

Loyalty to our species might be a better defence. “Buy hand-made, save a human!” 
sounds like a plausible rallying-cry, or at least a marketing slogan. The past is not 
always a reliable guide to the exponential future, but it is a good place to start, and 
unfortunately it does not augur well for appeals to loyalty. In the late 1960s, Britain 
was feeling queasy as the Empire dissolved and Germany's economic power was 
returning. The “I'm backing Britain” campaign started in December 1967, trying to get 
British people to buy domestically manufactured products instead of imports. It fizzled 
out within a few months . [cclxxvii] 

Car manufacturing has long been symbolic of a nation's manufacturing virility. In the 
1950s, Britain was the world's second-largest car manufacturer after the US, but in the 
1960s its designs and build quality fell behind first its European rivals, and then the 
Japanese. Despite repeated appeals to buy British, sales declined and in 1975 the 
remaining national manufacturer, British Leyland, was nationalised. It never recovered, 
and Britain is now home to none of the major global car brands. (Fortunately it has 
many innovative and thriving automotive design and component businesses, and it 
manufactures in record numbers for foreign brands.) 

Appealing to people to buy handmade items out of loyalty to one's species may not have 
a huge economic impact if machine-made items are better quality and much, much 
cheaper. And in a world of falling employment, most people are going to have to buy as 
efficiently as they can. 

The third reason for buying from humans could be summarised by the phrase “artisanal 
variation”. We like antiques because the patina of age gives them personality: each one 
is unique. The same goes for the original work of an artist, even if it isn't a Vermeer or 

a Rubens. But for most people, this is the preserve of luxury items, a few select pieces 
which we keep on display. Most of our possessions are mass-produced because they 
are much cheaper, disposable, and we can afford a better lifestyle that way. 

We have seen this before, in the second half of the 19 th century. With the industrial 
revolution in full swing, William Morris helped found the Arts and Crafts movement to 
produce hand-made furnishings and decorations. His concern was to raise quality 
rather than to reduce unemployment, but in practice he ended up making expensive 
pieces which only the rich could afford, [cclxxviii] 

Some people may choose to buy goods and services from humans rather than machines 
for reasons of status. But by definition, this could only ever amount to a niche activity, 
and would not save most of us from unemployability. 

New jobs 

If machines are going to take a great many, perhaps most, of our existing jobs, can we 
create a host of new ones - perhaps whole new industries - to replace them? Those 
who think we can point out that many of the jobs we do today did not exist a hundred 
years ago. Our grandparents would not have understood what we mean by website 
builder, social media marketer, user experience designer, chief brand evangelist, and so 
on. Surely, the argument goes, all these new technologies we have been talking about 
will throw up many new types of jobs that we cannot imagine today. 

As the person probably most responsible for Google's self-driving cars, Sebastian 
Thrun is a man worth listening to on the subject. He is optimistic: “With the advent of 
new technologies, we’ve always created new jobs. I don’t know what these jobs will 
be, but I’m confident we will find them.” [cclxxix] 

Unfortunately, past experience is (again) not as encouraging as you might think. Gerald 
Huff is a software engineer working in Silicon Valley, ground zero of the developments 
we are talking about. Nervous about the prospect of technological unemployment, he 
carried out a comparative analysis of US occupations in 1914 and 2014. Using data 
from the US Department of Labour, [cclxxx] he discovered that 80% of the 2014 
occupations already existed in 1914. Furthermore, the numbers of people employed in 
the 20% of new occupations were modest, with only 10% of the working population 
engaged in them. The US economy is much bigger today than it was in 1914, and 
employs far more people, but the occupations are not new. 

Of course, those of us who argue that it is different this time cannot rely on the historical 

precedent. It might be different this time in that vast swathes of new jobs will be 
created - including jobs for averagely-skilled people, not just relatively high-skill jobs 
like social media marketing. But those who argue that we are falling for the Luddite 
fallacy cannot argue that history points to everybody getting new types of jobs which are 
more interesting and safer after a period of adjustment. It doesn't. 

If we were to create a host of new jobs, what might they be? Maybe some of us will 
become dream wranglers, guiding each other toward fluency in lucid dreaming. Others 
may become emotion coaches, helping each other to overcome depression, anxiety, and 
frustration. Maybe there will be jobs for which we have no words today, because the 
technology has not yet evolved to allow them to come into being. 

It's not hard to imagine that virtual reality will create a lot of new jobs. If it is addictive 
as enthusiasts think it will be, many people will spend a great deal of their time - 
perhaps the majority of it - in VR worlds. In that case there will be huge demand for 
new and better imaginary or simulated worlds to inhabit, and that means jobs. 

But does it mean jobs for humans? Although the credit list for the latest superhero 
blockbuster stretches all the way around the block as it names everyone involved in 
rotoscoping and compositing the hyper-realistic armies of aliens, the latest CGI 
technology also makes it possible for two teenagers with a mobile phone to make a film 
which gains theatrical distribution. Their increasingly powerful software and hardware 
allows Hollywood directors to conjure visual worlds of such compelling complexity 
that their predecessors would rub their eyes in disbelief, but it also allows huge 
quantities of immersive content to be developed by skeleton crews. There will 
probably always be an elite of directors who are highly paid to push the boundaries of 
what can be imagined and what can be created, but software will do more and more of 
the heavy lifting in VR production. 

Not for the first time, the games industry shows what is possible. A game called “No 
Man’s Sky” was announced in 2014 which conjures far more imaginary worlds than you 
could visit in a lifetime purely by the operation of algorithms and random number 
generators. You boldly go where no programmer or designer has gone before. [cclxxxi] 


It is surprising how many smart people think that education is the answer to automation 
by machine intelligence. Microsoft CEO Satya Nadella said in January 2016: “I feel the 
right emphasis is on skills, rather than worrying too much about the jobs [which] will be 
lost. We will have to spend the money to educate our people, not just children but also 

people mid-career so they can find new jobs.” [cclxxxii] 

Massive Open Online Courses, or MOOCs, are promoted as the way we will all re¬ 
train for a new job each time a machine takes our old one. MOOCs are important, and 
along with flipped lessons, competency-based learning, and the use of Big Data, they 
will improve the quality of education, and make excellent learning opportunities 
available to all. (With flipped lessons, students watch a video of a lecture for 
homework, and then put what they have been told into practice in the classroom. The 
teacher acts as coach and mentor, a more interactive role than lecturing. 
Competency-based learning requires students to have mastered a skill or a lesson before 
they move on the next one; students within a class may progress at different speeds. Big 
data enables students and teachers to understand how well the learning process is going, 
and where extra support is needed.) 

Exciting and powerful as these techniques are, they won't protect us from technological 
unemployment. We have seen that machines are increasingly capable of performing 
many of the tasks currently carried out by highly educated, highly paid people. The 
machines aren't just coming for the jobs of bricklayers; they're coming for the jobs of 
surgeons and lawyers too. 

There is a very important postscript to these remarks about education. If we make it 
through successfully to the new world in which many or most of us are permanently and 
irrevocably unemployed, then education will be more important, not less. We will need 
good education to take advantage of our leisured lives even more than we did to survive 
our working lives. 


If machines take over the jobs that are repetitive, humans will look to do things that 
require creativity, intuition, and pursue counter-intuitive paths. One job title which fits 
that description is entrepreneur. 

In my experience there are two types of entrepreneur. Both are resourceful, determined, 
and usually of above-average intelligence. The first and most common type is someone 
who works in an organisation which is doing something poorly. They notice this, and 
decide to offer a better version. They utilise essential skills and industry know-how 
acquired while working for the original organisation, and simply improve incrementally 
on what was being provided there. These people are talented and hard-working, but 
they also had the good fortune to be in the right place at the right time to spot the 
opportunity. If they had not been in that position they would have spent their careers 

working for other people, and because they are hard-working and bright they would 
probably have make a good fist of that. 

The second type is destined to be an entrepreneur whatever circumstances life drops 
them into. They will never be happy working for someone else. They envision 
themselves in a future world which looks impossible to anyone else, but they choose to 
believe it and by dint of sheer force of will they make that future a reality. They will 
walk through brick walls to make it happen, and will probably be bankrupt more than 
once. They are charming, astonishingly energetic, and often rather hard to be around. In 
the words of Linkedln founder Reid Hoffman, they are people who will happily throw 
themselves off a cliff and assemble an aeroplane on the way down. [cclxxxiii] 

Both types of entrepreneur are rare, and especially the second kind - which may be a 
good thing for the rest of us. In any case, this is probably not an occupation that is going 
to save large numbers of people from technological unemployment. The other thing to 
remember about entrepreneurship as a career is that most startups fail. 


After all these apparently gloomy prognostications, let's close this chapter on a more 
optimistic note. There is one profession which can probably never be automated until 
the arrival of an artificial general intelligence which is also fully conscious. That 
profession is art, and to understand why, it is important to distinguish between art and 

Creativity is the use of imagination to create something original. Imagination is the 
faculty of having original ideas, and there seems to be no reason why that requires a 
conscious mind to be at work. Creativity can simply be the act of combining two 
existing ideas (perhaps from different domains of expertise) in a novel way. 

The eminent 19 th -century chemist August Kekule solved the riddle of the molecular 
structure of benzene while day-dreaming, gazing into a fire. [cclxxxiv] True, he had 
spent a long time before that pondering the problem, but according to his own account, 
his conscious mind was definitely not at work when the creative spark ignited. You 
might argue that Kekule's sub-conscious was the originator of the insight, and that a sub¬ 
conscious can only exist where there is consciousness, but that seems to me an assertion 
that needs proving. 

Can computers be creative? In mid-2015, Google researchers installed a feedback loop 

in an image recognition neural network, and the result was a series of fabulously 
hallucinogenic images, [cclxxxv] To deny that they were creative is to distort the 
meaning of the word. 

Art is something different. Admittedly, this is a personal definition, and perhaps not 
everyone would agree, but surely art involves the application of creativity to express 
something of personal importance to the artist. It might be beauty, an emotion, or a 
profound insight into what it means to be human. (If that disqualifies a good deal of what 
is currently sold under the banner of art, then so be it - in fact, three cheers.) 

To say something about your own experience clearly requires you to have had some 
experience, and that requires consciousness. Therefore, until a conscious artificial 
general intelligence (AGI) arrives, AIs can be creative but not artistic. This in turn 
means that while Donna Tartt and Kazuo Ishiguro are probably OK for a few decades, 
today’s successful genre writers who use stables of assistants to churn out several crime 
and romance novels each year have chosen the timing of their careers expertly, and their 
publishers had better find something different to do. 

In April 2016, researchers from Microsoft, a Dutch university and two Dutch art 
galleries created an AI which analysed the way Rembrandt painted. It identified enough 
of his techniques and mannerisms to enable it to produce paintings in exactly his style - 
better than any human forger could. They had it design a new picture in Rembrandt’s 
style, of a subject Rembrandt had never worked on, and 3D printed it to capture the old 
master’s technique in three dimensions. Because the machine recognises patterns better 
than humans can, it may well teach us interesting new things about the way Rembrandt 
created his masterpieces. But it is not producing art. [cclxxxvi] 

3.11- What's the problem? 

The Star Trek economy 

It is often said that science fiction tells us more about the present than it does about the 
future. Most science fiction writers are not actually trying to predict the future, although 
they may go to considerable lengths to try to make the worlds they create seem 
plausible. Generally, they are just trying to tell an entertaining story, or maybe use the 
opportunity that the genre offers to explore something about the fundamental nature of 
our lives. (At its best, science fiction is philosophy in fancy dress.) 

But intentionally or otherwise, science fiction does a very important job for all of us 
when we think about the future: it provides us with metaphors and scenarios. Many of 
the most popular science fiction stories present dystopian scenarios: think Terminator, 
Blade Runner, 1984, Brave New World, and so on. But there are also positive 
scenarios, and one of the most popular ones is Star Trek. 

Set in the 24 th century, Star Trek presents a world of immense possibility, of interstellar 
travel, adventure, and split infinitives. And a world without money or poverty. In the 
1996 movie “Star Trek: First Contact”, Captain Jean-Luc Picard explains that “Money 
doesn't exist in the 24th century. The acquisition of wealth is no longer the driving force 
in our lives. We work to better ourselves and the rest of humanity." 

This was not a feature of the original TV series, in which there were quite a few 
mentions of money, and systems of credit. But before he died, Gene Roddenberry 
stipulated that there was to be no money in the Federation, [cclxxxvii] 

Although there is no money, the people in the later Star Trek stories do compete with 
each other - for prestige, for approval, for increased responsibility and for career 
advancement. One of the things that makes James Tiberius Kirk an outstanding starfleet 
commander is his fiercely competitive nature. He operates in a profoundly meritocratic 
environment, and will sacrifice a great deal to win. 

This is not new. Men and women have always competed for pre-eminence within their 
tribes and societies, and we are continually applying our ingenuity to work out new 
ways to do so. Mediaeval knights risked life and limb for honour and glory, and their 
descendants fought for national self-determination. Today, many people expend 
considerable sweat and tears - if less blood - to demonstrate their prowess in writing 

elegant open source software, or edit Wikipedia pages. 

Money is not required in Star Trek's United Federation of Planets because energy has 
become essentially free, and products can be manufactured in so-called Replicators, 
devices which create useful (including edible) objects out of whatever matter is 

Another popular science fiction series with a broadly optimistic (if darkly humorous) 
outlook is the late Iain M. Banks' “Culture” books, set in a distant future when a 
technologically advanced humanity has colonised swathes of the galaxy, and enjoys 
mostly peaceful relations with a host of alien civilisations. The humans are kept 
company and aided by vastly superior and extraordinarily indulgent machine 
intelligences, and they lead lives of perpetual indulgence. As Banks put it in a 2012 
interview, “It is my vision of what you do when you are in a post-scarcity society, you 
can completely indulge yourself. The Culture has no unemployment problem, no one has 
to work, so all work is a form of play.” [cclxxxviii] 


The Star Trek economy is the post-scarcity economy, the economy of radical 
abundance. In their 2012 book “Abundance: the future is better than you think”, Peter 
Diamandis and Stephen Kotler argue that this world is within reach in the not-too- 
distant future, thanks largely to the exponential improvement in technology. 

Financial Times columnist Martin Wolf urged that we should “enslave the robots and 
free the poor”. [cclxxxix] and who would not welcome such an outcome? Perhaps if we 
play our cards right, automation by machine intelligence will simply mean that we 
humans get to spend our long and healthy lives playing, learning, enjoying each other's 
company, having adventures and fun. 

Of course, life is rarely so simple or so easy. In chapter 5 we will explore some of the 
challenges and hurdles to be overcome. But let's pause for a moment, and beguile 
ourselves with Richard Brautigan's poetic wish, recorded back in 1967. 

“I like to think 

(it has to be!) 

of a cybernetic ecology 

where we are free of our labors 

and joined back to nature, 

returned to our mammal 

brothers and sisters, 

and all watched over 

by machines of loving grace.” [ccxc] 

3.12 - Conclusion: yes, it’s different this time 

It's time to answer the question: is it really different this time? Will machine 
intelligence automate most human jobs within the next few decades, and leave a large 
minority of people - perhaps a majority - unable to gain paid employment? 

It seems to me that you have to accept that this proposition is at least possible if you 
admit the following three premises: 

1. It is possible to automate the cognitive and manual tasks that we carry out to do our 

2. Machine intelligence is approaching or overtaking our ability to ingest, process and 
pass on data presented in visual form and in natural language. 

3. Machine intelligence is improving at an exponential rate. This rate may or may not 
slow a little in the coming years, but it will continue to be very fast. 

No doubt it is still possible to reject one or more of these premises, but for me, the 
evidence assembled in this chapter makes that hard. 

Counter- arguments 

The main argument against the proposition that machines will create widespread 
structural unemployment is that it hasn’t happened in the past. In other words, the 
proposition is the Luddite fallacy. This is a weak argument at best - akin to saying that 
we have never sent a person to Mars so we never will. It is also partly false: during the 
Engels Pause in the first half of the 19 th century unemployment rose, and labour’s share 
of national income fell, in the UK at least. But the main point is that the whole question 
is whether it will be different this time. 

People sometimes argue that humans will not become unemployable because there is an 
inexhaustible well of human wants and needs to fulfil. This doesn’t help much if 
machines can fulfil any need better, faster and cheaper: it will always be economically 
compelling to build a machine to do the job. 

The best argument against the proposition is that humans possess skills which machines 
will never replicate - or at least, not for many decades. This is inevitably a judgement 
call, since at the moment, no-one can be certain. My judgement, based on the evidence 
set out above, is that the skills for which we get paid will be acquired by machines in 

the next two to four decades. 

Raising awareness 

2015 was an important year for artificial intelligence. It was the year when our media 
caught on to the idea that AI presents enormous opportunity and enormous risk. This 
was thanks in no small part to the publication the previous year of Nick Bostrom's book 
“Superintelligence”. It was also the year when cutting-edge AI systems used deep 
learning and other techniques to demonstrate human-level capabilities in image 
recognition, speech recognition and natural language processing. In hindsight, 2015 may 
well be seen as a tipping point. 

Machines don't have to make everybody unemployed to bring about an economic 
singularity. If a majority of people - or even just a large minority - can never get hired 
again, we will need a different type of economy. 

Furthermore, we don't have to be absolutely certain of this outcome to make it 
worthwhile to monitor developments and make contingency plans. After all, seeing past 
the event horizon of a singularity is hard. 

As we have seen, and as we will explore in more detail in chapters 5 and 6, the 
outcome can be fantastic or terrible. To a large extent, it is up to us. If we can't achieve 
the positive, then perhaps we deserve (as Anders Sandberg put it in chapter 3.4) to be 
the boot loader for the digital superintelligence rather than its mitochondria, [ccxci] 

4. - A timeline 

4.1- Un-forecasts 

Three snapshots of a positive scenario 

This chapter offers three snapshots of a possible future, one each for 2021, 2031 and 
2041. Their purpose is to make the possibility of technological automation seem more 
real and less academic. 

In each one there is a very brief description of the level of automation in a number of 
industries, and a summary of the impact of that automation on society. Together they 
depict a positive scenario, with a concluding vision of an economy of radical 
abundance which has been achieved without massive social dislocation. Why choose a 
positive scenario? Why not: it's the outcome we should be aiming for. 

Before we start, there is an important caveat. 

Unpredictable yet inevitable 

We know that all forecasts are wrong. The only things we don't know are by how much, 
and in what direction. The future generally turns out to be not only different to what we 
expect, but also much stranger. Cast your mind back to 2005. Pretty much everyone 
thought that cellphones would continue to get smaller, and Facebook was limited to a 
few thousand universities and schools. Today, just a decade later, larger smartphones 
are a bit of a thing, and Facebook's valuation has overtaken that of Walmart, the world's 
largest shopkeeper. [ccxcii] Trying to predict how the world will look in 2031 is like 
trying to predict the weather on Saturday two months from now. There are just too many 

And yet in hindsight what happens appears not only natural, but almost inevitable. 

The smartphone is a good example. Pretty much nobody suggested thirty years ago that 
we would all have telephones in our pockets which would contain powerful artificial 
intelligences, and which would only occasionally be used for making phone calls. After 
all, at the time a mobile phone was a fairly hefty device, the size of a small dog. But 

now that it has happened it seems obvious, logical, and perhaps even inevitable. 

Here’s why. We humans are highly social animals, and our social habits are facilitated 
by language. Because we have language we can communicate complicated ideas, 
suggestions and instructions: we can work together in teams and organise; we can 
defend ourselves against lions and hostile tribes, we can hunt and kill mammoths, 
produce economic surpluses, and develop technologies. 

It is often said that no species is more savage and more violent than humans. This is no 
more true than the claim that Americans are more violent than other nationalities 
because their murder rate is higher than other developed countries. The only reason 
why humans kill more than other species is that we have more and better weapons. 

Humans live cheek by jowl in cities containing millions. This is remarkable: no other 
carnivorous species can assemble more than a few dozen of its members in a limited 
space without them killing each other in rivalry for food, sex, or social dominance. 

Other species lack our sophisticated ways to communicate and collaborate. Our bigger 
brains allow us to establish laws and cultural norms which govern the way we interact. 
We make up stories and agree to believe in them collectively, regardless of whether 
there is any evidence for them. These stories - about abstract concepts like gods and 
kingship, nations and ideologies, money and art - give us powerful reasons to cooperate 
and work together, even to die together. 

Non-human primates spend hours every day grooming the other members of their tribe to 
reassure them that they will not sink their teeth and nails into them. It works, but it is 
inefficient, and means they cannot readily add new members to their tribe. Humans, by 
contrast, can walk past complete strangers on a crowded street without a second 
thought. Our superpowers are communication, and our capacity to sustain mutual belief 
in things for which we have no evidence. It is thanks to these abilities that we control 
the fate of this planet and every species on it. 

(This means that the old cliche that our dominance is based on our capacity for rational 
thought is - unlike most cliches - untrue.) 

So although it wasn't - and couldn't have been - predicted in advance, in hindsight it is 
entirely logical that our most powerful technology, artificial intelligence, would first 
become available to most of us in the form of a communication device. 

The way the economic singularity unfolds will probably be like that. Our attempts to 
forecast the impact of technological unemployment - assuming it arrives - will 

probably look absurd in hindsight. But when we get there, the outcome will seem not 
only natural, but perhaps even inevitable. 

Not forecasts 

I am labouring this point because I want to be clear. The descriptions of a possible 
future that follow are not predictions. The only thing I am confident of is that the future 
will not be like this. 

Instead, these timelines are intended to serve two functions. First, as I said above, they 
are a rhetorical device. The arguments in chapter 3.10 that machines will automate our 
jobs away have been either abstract or fragmentary, and as such, some readers may find 
them implausible. I'm hoping that the timelines will help make the possible future of an 
economic singularity seem less academic, less theoretical, and more real. 

Secondly, drawing up timelines like these may in some small way help us to construct a 
valuable body of scenarios. Even when we know the future is unpredictable, it is still 
worth making plans. There is good sense in the old cliche that failing to plan is 
planning to fail. If you have a plan, you may not achieve it, but if you have no plan, you 
most certainly won't. 

In a complex environment, scenario development is a valuable part of the planning 
process. None of the scenarios will come true in their entirety, and many will be 
completely off the mark. But parts of some of them may approximate some of the 
outcome. Thinking through how we would respond to a sufficient number of carefully 
thought-out scenarios could well help us to react more quickly when we see the 
beginnings of what we believe to be a dangerous trend. 

Super un-forecasting 

The art of constructing a useful scenario is the same as that of forecasting, which has 
been extensively studied by Canadian political scientist Philip Tetlock, co-author of the 
book “Superforecasting: the art and science of prediction.” He has found that the best 
forecasters share a number of traits. First, they treat their views about what will happen 
as hypotheses, not firm beliefs. If the evidence changes, they change their hypothesis. 

Secondly, they look for numerical data. Now we all know that there are lies, damned 
lies and statistics, and that data is often used in public debate in the same way that a 
drunk uses a lamp-post: more for support than for illumination. But used carefully and 

honestly, data is our friend. It is after all the root of the scientific revolution that has 
lifted most of our species out of poverty and squalor. 

Thirdly, they look for context. He cites the example of guests at a wedding, admiring 
the beauty and grace of the bride and the dashing good looks of the bridegroom, and 
assuring each other that they will share a long and happy life together. The super¬ 
forecaster is a contrarian, noting that around half of all marriages fail, and that the 
failure rate increases with second and third marriages, especially when one or other 
partner has a history of infidelity, as with the happy couple today. If she is a tactful 
super-forecaster, she keeps these thoughts to herself. 

Ironically, super-forecasters are often not the people who get listened to in discussions 
about the future. We tend to pay more attention to those who speak most confidently, 
and offer clarity and certainty. People who equivocate and offer measured suggestions 
often don't cut through the noise. 

So here goes, with the equivocation minimised. 


1. Transportation. Numerous cities around the world are experimenting with self¬ 
driving cars, but very few are so bold as to omit human drivers altogether. Google's 
cars can now handle snow and heavy rain, and can distinguish between a pedestrian 
waving at a friend across the street and a policeman instructing the car to stop. Their 
sensors are still expensive, but they are getting cheaper quickly. Aspects of self-driving 
technology are becoming widespread, including intelligent cruise control and assisted 
parking. Many long-distance commuters let their cars do much of their driving, although 
watching TV while driving is still illegal. 

2. Manufacturing. Industrial robots are getting cheaper, and much easier to 
programme to undertake new tasks. Manufacturers find that the choice between 
employing another person or buying a new robot is a close thing. 

3. Agriculture. Farmers are experimenting with robots for both crops and animal 
husbandry. On a growing number of farms with high-value crops, small wheeled 
devices patrol rows of vegetables, interrogating plants which don't appear to be healthy 
specimens, and eliminating weeds with targeted jets of ecologically-sound hot water. 
Cattle are entirely content to be milked by robots, so the declining population of farm 
workers no longer has to get up before daybreak every day. 

4. Retail. The shift towards purchasing goods and services online continues, and there 
is growing automation within shops. In some supermarkets, shoppers no longer have to 
unload and re-load their trolleys: the goods are scanned while still inside their baskets. 
Fewer attendants are required in the checkout area. In fast food outlets, so-called 
“McJobs” are disappearing as burgers and sandwiches are assembled and presented to 
customers without being handled by a human. 

5. Construction. Although some developers are experimenting with pre-fabricated 
units, most of the cost of a construction project is generated by the variability of 
conditions on-site, including the foundations. Robots which can handle this 
unpredictability are still too expensive to replace human construction workers. There 
are experiments with exoskeletons for construction workers, but these are still 

6. Technology. Firms are fighting to recruit and retain machine learning experts; the 

salaries and bonuses offered were previously unknown outside financial services and 
professional sports. Sales of wearables are growing, and the successors to Google 
Glass are out-selling smart watches. 

7. Utilities. Water companies and power generation and transmission firms are 
building out fleets of tiny robots and drones which patrol pipes and transmission lines, 
looking for early warning signs of failure. 

8. Finance. Retail banking is mostly automated and web-based, and consumer 
feedback on the quality of service is improving. Wealthy people now get some of their 
investment advice directly from automated systems, but human investment advisers still 
serve most of the market. In corporate finance, human advisers show no signs of being 
replaced, although their back office systems are heavily automated. 

9. Call centres. Enquiry handling that was offshored to India and then repatriated to 
home countries is now being offshored again - this time to machines housed in cold 
climates where the cost of keeping the servers cool are lower. 

10. Media and the arts. The market for virtual reality apps and shows is booming as 
Oculus Rift, Meta, and their competitors create demand for latency-free, high resolution 
content. As usual, porn and sport look like being the killer apps, but there are 
unexpected hits too, such as “how-to” shows about parenting and relationship 

11. Management. Little change. 

12. Professions. The tedious jobs which traditionally provided training wheels for 
accountants and lawyers (“ticking and bashing” for auditors and “discovery” for 
litigators) are increasingly being handled by machines. Optimists - and sceptics about 
technological unemployment - point out that the amount of work for trainee 
professionals has actually increased, as whole categories of previously uneconomic 
jobs have become possible, and the machines still need training on each new data set. 
But more thoughtful practitioners are writing articles in their trade magazines asking 
how long that will continue, and therefore how tomorrow's partners can learn their 

13. Medical. In two Scandinavian countries and a handful of US states, AIs are 
ingesting data sent by patients from their smartphones and carrying out triage. 

Sometimes they respond with simple diagnoses and treatment recommendations, 
sometimes they pass the enquiry to a human doctor. Medical professionals and 

regulators elsewhere are highly critical of these experiments, but the outcome data is 
impressive. Hospitals in Japan are using robot nurses to great effect, but these are also 
resisted elsewhere. Pharmaceuticals designed to raise the IQ of adults are in clinical 

14. Education. Teachers everywhere brandish the empirical evidence that the primary 
determination of educational outcome is the quality of the teaching. Some teachers are 
embracing new technologies enthusiastically, especially in competitive environments 
such as the UK's private school system. Others are resistant. 

15. Government. There is a worldwide drive to get most government “services” 
delivered online and cheaper. 


There is a great deal of discussion in the media about whether automation will lead to 
technological unemployment. A growing number of people think that it will, but they 
are outnumbered by those, including many prominent economists, who continue to deny 


1. Transportation. Most long-distance trucks are now capable of operating 
autonomously, and many operate routinely without a human on board. In some 
jurisdictions there are roads which are off-limits to human drivers. Most cars are still 
privately owned, but many people are experimenting with communally-owned vehicles, 
which enjoy free parking in most cities. Insurance premiums have plummeted, and fears 
about self-driving cars being routinely hacked have not been realised. A vocal minority 
of citizens (which, to general surprise, comprises equal numbers of men and women) 
are scathing about this arrangement, dubbing the communal cars THEMs, (“tedious 
horizontal elevator machines”). 

Many urban deliveries of fast food and small parcels in major cities are now carried out 
by autonomous drones, operating within their own designated level of airspace. 
Sometimes the last mile of a delivery is carried out by autonomous wheeled containers. 
Teenagers delight in “bot-tipping”, but with all the cameras and other sensory 
equipment protecting the bots, it is a risky pastime. 

2. Manufacturing. Many large factories and warehouses are dark: no light is required 
because no humans work there. People are becoming a rarity in smaller sites too. 

3D printing has advanced less quickly than many expected, as it remained more 
expensive than mass production. But it is common in niche applications, like urgently 
required motor parts. 

3. Agriculture. Farmers are moving heavily into leisure services, as their families and 
staff are losing their roles to robots. 

4. Retail. Online shopping reaches 75% of all retail purchases, with a small but 
growing number of items being 3D printed domestically or in neighbourhood facilities, 
often with an element of customisation by the consumer. Human shop assistants are 
starting to be replaced by robots, except in high-margin sectors where they help create 
an experience rather than simply facilitating straightforward transactions. 

5. Construction. Human supervision is still the norm for laying foundations, but pre¬ 
fabricated (often 3D-printed) walls, roofs and whole building units are becoming 
common. Robot labour and humans in exoskeletons are increasingly used to assemble 
them. Drones populate the air above construction sites, tracking progress and enabling 

real-time adjustments to plans and activities. 

6. Technology. The first “inside-ables” are appearing, and have been made 
fashionable by Lord Beckham, the football and fashion magnate. The Internet of Things 
has materialised, with everyone receiving messages continuously from thousands of 
sensors and devices implanted in vehicles, roads, trees, buildings, etc. Fortunately, the 
messages are intermediated by personal digital assistants, which have acquired the 
generic name of “Friends”, but whose owners often endow them with pet names. 

New types of relationship and etiquette are evolving to govern how people interact with 
their own and other peoples' “Friends”, and what personalities the Friends should 
present. Brand loyalty to the companies which provide the best Friends software is 

There is lively debate about the best ways to communicate with their “Friends” and 
other computers. Most people communicate with them by muttering into implanted 
microphones, but millions of people are also learning how to use one-handed keyboards 
which liberates them from traditional keyboards at times when voice is inappropriate. 
Some believe these new keyboards will be quickly superseded by Brain-Computer 
Interfaces (BCI), but this has made less progress than its early enthusiasts expected. 

A growing amount of entertainment and personal interaction is mediated through virtual 
reality. It is increasingly rare to see an adolescent in public outside school hours. 

Polls suggest that most people now think that artificial general intelligence (AGI - 
machines which equal or surpass human cognition in all domains) is a serious 
possibility within a generation or two. Significant expenditure is flowing into research 
on how to make sure the outcome is positive, and the moral and religious implications 
are hotly debated. 

7. Utilities. In many organisations, most operations are now automated. The main role 
of humans in these organisations is testing security arrangements. Several hundred 
people died in two significant hacking incidents - one in the US and one in Europe. 

This has prompted huge investment in upgraded security arrangements. In another high- 
profile incident, AI management systems managed the disaster containment and recovery 
process flawlessly, and much faster than humans could have. 

8. Finance. Retail banking is now fully automated, and investment advice is going the 
same way. Corporate financiers are in retreat, and their previously stratospheric 
incomes have fallen sharply. 

9. Call centres. Almost no humans now work in call centres. 

10. Media and the arts. All major movies made by Hollywood and Bollywood are 
now produced in VR, along with all major video games. To general surprise, levels of 
literacy - and indeed book sales - have not fallen. In a number of genre categories, 
especially romance and crime, the most popular books are written by AIs. 

Major sporting competitions have three strands: robots, augmented humans, and un¬ 
augmented humans. Audiences for the latter category are dwindling. 

Long-distance communication is massively improved by VR Skype. 

Dating sites have become surprisingly effective by requiring their members to provide 
clothing samples from which they extract data about their smells and their pheromones. 
The discovery that relationship outcomes correlate closely with these data have slashed 
divorce rates. 

11. Management. The ranks of middle management are thinning out. Shareholders are 
investing heavily in Distributed Autonomous Corporations (DACs), firms consisting of 
unsupervised AIs which create new business models and strategies and transact with 
other firms without any humans in the loop. 

12. Professions. Partners in law firms and accountancy firms are working shorter 
hours. Human intakes to these firms are dwindling. Most criminal law cases now 
relate to digital crime: it is more lucrative than physical crime, and easier to avoid 

13. Medical. Opposition to the smartphone medical revolution has collapsed inmost 
countries, and most people obtain diagnoses and routine health check-ups from their 
“Friends” several times a week. Automated nurses are becoming increasingly popular, 
especially in elder care. 

Several powerful genetic manipulation technologies are now proved beyond reasonable 
doubt to be effective, but backed by public unease, regulators continue to hold up their 
deployment. Cognitive enhancement pharmaceuticals are available in some countries 
under highly regulated circumstances, but are proving less effective than expected. 

There are persistent rumours that they are deliberately being engineered that way. 

Ageing is coming to be seen as an enemy which can be defeated. 

14. Education. Data on learning outcomes is steamrollering teachers' resistance to 
new approaches. Customised learning plans based on continuous data crunching are 
becoming the norm. Teachers are becoming coaches and mentors rather than 
instructors. Some schools are experimenting with classroom AIs. 

15. Government. There is growing pressure to reduce the numbers of politicians and 
civil servants, as more and more government services are automated. Many 
jurisdictions are debating the merits of using technology to enable direct democracy, 
which is being pioneered by Switzerland. Most people are sceptical, fearing the 
tyranny of the temporary majority. Policemen in most countries record all interactions 
with members of the public, and public satisfaction levels with them are generally 

16. Charities. Non-profit organisations are enjoying a surge, thanks to an influx of 
talent as capable people can't find work elsewhere. 

Discussion and experimentation 

The majority of people in most countries now believe that an economic singularity is 
coming and that a universal basic income (UBI) will be needed, as before long a very 
large minority of citizens will be permanently unemployed. Experiments in numerous 
cities around the world, and a couple of country-scale experiments demonstrate that 
most people don't succumb to drugs or despair, although a significant minority does, and 
needs help. There is vigorous debate about how to pay for the UBI. 


1. Transportation. Humans very rarely drive vehicles on public roads, and few 
commercial vehicles have human attendants. Young people no longer take driving 
tests. Motor sports are mostly competitions between self-driving cars. Congestion and 
parking are no longer a problem. The population of cars has declined dramatically as 
they are used far more efficiently, and the automotive industry has contracted. Large 
numbers of dependent businesses (and jobs) are disappearing too, including repair 
shops and insurance brokers. 

2. Manufacturing. Almost all factories and warehouses are dark. 3D printing is 
beginning to look competitive with some forms of mass production. 

3. Agriculture. Robots do most farm work. 

Some countries have large communally-owned agricultural processing concerns which 
send out meal ingredients on drones in a service described as Netflix for food. 

4. Retail. Most items are now bought online, and around half of all products sold at 
retail are 3D printed. Retail outlets on High Streets and city centres are mostly 
experiential rather than transactional, and mostly staffed by AIs and robots. 

5. Construction. Robots now carry out most of the work on construction sites. 

6. Technology. Since AI provides a large proportion of the value in most products and 
services, there is a major concentration of capital and wealth in the hands of 
shareholders and key employees in this sector. Its foremost talent is now applied to 
developing artificial general intelligence and making sure that it is safe for humans. The 
Internet of Things is all-pervasive, and the environment appears intelligent. 

The companies that provide “Friends” have been obliged to make them open-source. 
Friends are so critical to everyone’s lives that being restricted to any one company’s 
walled garden was unacceptable. 

7. Utilities. Overwhelmingly automated. 

8. Finance. Overwhelmingly automated. 

9. Call centres. Unchanged. 

10. Media and the arts. In sports, robot competitions now generate larger audiences 
than their human counterparts. The International Olympics Committee de-lists the 
human versions of around half of all sports. 

Haptic body suits combined with VR headsets now provide truly immersive virtual 
environments. Counselling (by AIs) is required by a section of the population who 
struggle to maintain the distinction in their minds between reality and VR. 

To general surprise, people still read books, but they are very different products now, 
with holographic illustrations, and often with several alternative story lines developed 
by their AI authors, which readers can choose between. 

Dating sites are now mostly accessed by personal digital assistants. “My Friend likes 
your Friend” has become a standard opener. 

11. Management. Many companies now consist of just a few strategists, whose main 
role is to forecast the optimal business model for the next financial quarter, but they are 
struggling to keep up with their AI advisers. 

12. Professions. Accountancy and the law are largely automated. 

13. Medical. Demand for human doctors is dwindling and professional nursing has 
been almost entirely automated. Everyone in developed economies has their health 
monitored continuously by their “Friends”. Most people spend a certain amount of time 
each week visiting family, friends and neighbours who are unwell, just to converse. 

Sick and disabled people are greatly comforted by their relationships with talking AI 
companions, some resembling humans, others resembling animals. Significant funds are 
now allocated to radical age extension research, and there is talk of “longevity escape 
velocity” being within reach - the point when each year, science adds a year to your life 
expectancy. Most forms of disability are now offset by implants and exoskeletons, and 
cognitive enhancements through pharmaceuticals and brain-computer interface 
techniques are showing considerable promise. 

14. Education. The sector has ballooned, with many people now regarding it as 
recreation rather than work. Most education is provided by AIs. 

15. Government. Safeguards have now been found to enable direct democracy to be 

implemented in many areas. Professional politicians are now rare. 

Radical abundance 

Unemployment has passed 50% in most developed countries. Some form of universal 
basic income or negative taxation is in place everywhere, and most people think the 
economic singularity has happened. 

Nobody hates their job. People only do work that they enjoy, and most people could not 
find paid jobs even if they wanted to. Everyone receives a basic income from the state 
or a non-centralised public organisation, and there is no stigma attached to being 
unemployed, or partially employed. In most countries the UBI was funded initially by 
taxes levied on the minority of wealthy people who own most of the productive capital 
in the economy, and in particular on those who own the AI infrastructure. 

In many countries, some of these elites have agreed to transfer the productive assets into 
communal ownership, either controlled by the state or by decentralised networks 
operated using blockchain technology (see chapter 6). Those who do this enjoy the sort 
of popularity previously reserved for film and sports stars. 

Some countries mandated these transfers early on by effectively nationalising the assets 
within their legislative reach, but most retreated from this approach when they realised 
that their economies were stagnating, as many of their most innovative and energetic 
people emigrated. Worldwide, the idea is gaining ground that private ownership of key 
productive assets is distasteful. Most people do not see it as morally wrong, and don't 
want it to be made illegal, but it is often likened to smoking in the presence of non- 
smokers. This applies particularly to the ownership of facilities which manufacture 
basic human needs, like food and clothing, and to the ownership of organisations which 
develop the most essential technology - the technology which adds most of the value in 
every industry sector: artificial intelligence. 

The gap in income and wealth between rich and poor countries has closed 
dramatically. This happened in part thanks to a substantial transfer of assets from the 
West to the rest, but mostly thanks to the adoption of effective economic policies, the 
eradication of corruption, and the benign impact of technology in the poorer countries. 

AIs and robots produce most of the goods and services people need in great abundance 
and at very low cost. Many products are 3D printed close to where they are consumed, 
so demand for commercial transportation services has plummeted. 

Demand for consumer travel is falling too, as immersive VR provides a close 
approximation of the experience. This takes the edge off the disappointment many 
people thought they would feel about not being able to increase their income by working 
harder in order to buy more luxury goods. Most people accept that they will never own 
a beach-front property with palm trees shading the white sand, but they can spend as 
much time there as they like in convincing VR. 

Another concern which has been allayed is that life without work would deprive the 
majority of people of a sense of meaning in their lives. Just as amateur artists were 
always happy to paint despite knowing that they could never equal the output of an old 
master like Vermeer, so people now are happy to play sport, write books, give lectures 
and design buildings in the knowledge that an AI could do any of those things better than 

Not everyone is at ease in this brave new world, however. Around 10% of the 
population in most countries suffers from a profound sense of frustration and loss, and 
either succumbs to drugs or indulges almost permanently in escapist VR entertainment. 

A wide range of experiments is under way around the world, finding ways to help these 
people join their friends and families in less destructive or limiting lifestyles. 

Governments and voters in a few countries resisted the economic singularity, seeing it 
as a de-humanising surrender to machine rule. Although they found economically viable 
alternatives at first, their citizens' standard of living quickly fell far behind. Several of 
these governments have now collapsed like the communist regimes of Eastern Europe in 
the early 1990s, and the others look set to follow - hopefully without violence. 

5. - The Challenges 

The point of this book so far has been to persuade you that within a few decades, it is 
likely that many people will be rendered unemployable by machine intelligence. If I 
have not wholly succeeded in that aim, then I hope you are at least prepared to accept 
that the possibility is serious enough that we should be thinking about the implications, 
and what to do about it if it happens. 

If I haven't even got you that far, then you're probably about to put this book down. If so, 
don't throw it away - you might want to come back to it when self-driving vehicles start 
to make serious impacts on the employment data. 

If I have made the case successfully - or if you were persuaded before we started - then 
welcome to the next stage of the journey. At the end of chapter 4 we saw a rosy 
scenario in which we are well on the way towards a new type of economy, and the 
transition has been smooth. 

Sadly, life is rarely smooth. There will be challenges. I anticipate five, and I think two 
of them might cause real harm if we are careless or unlucky. They are: economic 
contraction, distribution, meaning, allocation, and cohesion. Let's take a look at each of 
them in turn. 

5.1 - Economic contraction 

American union boss Walter Reuther recounts a story about a visit he made in the 1950s 
to a Ford manufacturing plant, where he saw an impressive array of robots assembling 
cars. The Ford executive who was showing him round asked how Reuther thought he 
would get the robots to pay union membership fees. Reuther replied that the bigger 
question was how the robots would buy cars. (The story is usually told with Henry 
Ford II playing the role of the company executive but it almost certainly 
wasn't, [ccxciii] ) 

The basic economic problem which this story is supposed to illustrate is that if nobody 
is earning any money then nobody can buy anything, and even those who do have money 
and resources can't sell anything. The economy grinds to a halt and everybody starves. 

Of course life is never as black-and-white as that. Economies don't go overnight from 
functioning tolerably well to complete collapse. Even catastrophic decline is less like 
falling off a cliff and more like tumbling down a slope, with pauses along the way as 
you hit ledges. But obviously, severe economic contraction is grim, and to be avoided 
if at all possible. 

If, as I have argued, machine intelligence renders more and more people unemployable, 
then other things being equal, the purchasing power previously exercised by those 
people will dry up. Their productive output will not be lost - it will just be provided 
by machines instead of humans. As demand falls but supply remains stable, prices will 
fall. At first, the falling prices may not be too much of a problem for firms and their 
owners, as the machines will be more efficient than the humans they replaced, and 
increasingly so, as they continue to improve at an exponential rate. But as more and 
more people become unemployed, the consequent fall in demand will overtake the price 
reductions enabled by greater efficiency. Economic contraction is pretty much 
inevitable, and it will get so serious that something will have to be done. 

But before policy makers are forced to take action to tackle economic contraction, they 
will be faced by a much more serious problem: what to do about all those people who 
no longer have a source of income? This is the distribution problem, which is seen by 
many as the most severe problem raised by the economic singularity. Tackling it 
successfully will also solve the problem of economic contraction, so we can move right 

5.2 - Distribution 

At the height of the Great Depression in the early 1930s, unemployment reached 25% of 
the working-age population, [ccxciv] Social security arrangements were primitive then, 
and developed societies were much poorer than they are today, so that level of 
joblessness was much harder on people than it is today, when parts of Europe have 
returned to similar levels overall. [ccxcv] with youth unemployment hitting 50% in some 
places, [ccxcvi] 

The worst levels of unemployment in developed countries today are found in 
Mediterranean countries like Greece and Spain, where family networks remain strong 
enough that sons and daughters can be supported for months or even years by fathers and 
mothers - and vice versa. There are escape valves, too, for the social pressure created 
by the situation. Economies further north are struggling less, and can absorb the 
energies and ambitions of many of the unemployed young people from the south. 

When self-driving vehicles and other forms of automation render people of all classes 
unemployed right across the developed world, these safety nets will no longer be 
available. Articulate, well-connected and forceful middle class professionals will be 
standing alongside professional drivers and factory workers, demanding that the state do 
something to protect them and their families. 

Universal Basic Income 

If and when societies reach the point where we have to admit that a significant 
proportion of the population will never work again - through no fault of their own - a 
mechanism will have to be found to keep those people alive. And not just scraping by 
on the poverty line: they will have to be provided with an income which allows at least 
the possibility of a decent life by the standards of the societies they live in. 

The answer is well-known, and fairly obvious: a universal basic income (UBI), 
available to all without condition; a living wage which is paid to all citizens simply 
because they are citizens. 

Probably the longest-standing organisation advocating UBI is the Basic Income Earth 
Network. BIEN was formed as long ago as 1986, and “Earth” replaced “European” in 
its name in 2004. BIEN defines UBI as “an income unconditionally granted to all on an 
individual basis, without means test or work requirement." UBI has also been called 

unconditional basic income, basic income, basic income guarantee (BIG), guaranteed 
annual income, and citizen's income. 

Proponents have argued for various levels of UBI, but in general they choose a level at 
or around the poverty level in the country of operation. This is partly because they don't 
think any more would be affordable, or politically acceptable, and partly to ward off 
criticisms that UBI would make people lazy and unproductive. As mentioned above, 
this will not be good enough if and when machine intelligence renders most people 
unable to work for a living. A modern developed society is not sustainable if a majority 
of its citizens are on the bread line. 

UBI is similar to but distinct from the concept of negative income tax (NIT), under 
which people earning less than a specified amount receive payments. The two systems 
can be set up to produce the same financial results, but they appeal to different 
economic and political instincts. UBI involves payments to people who really don't 
need them, while NIT could stigmatise recipients. 

The benefits claimed for UBI address issues which concern both the political left and 
right. Left-wing proponents see it as a mechanism to eradicate poverty and redress 
what they view as growing inequality within societies. They sometimes argue that it 
tackles the alleged gender pay gap, and redistributes income away from capital and 
towards labour. It has also been held out as a partial solution to the alleged 
generational theft whereby relatively wealthy pensioners are receiving income 
generated by taxes on young workers who have no assets, and who may not themselves 
receive similar benefits in later life because the welfare system looks increasingly 
unaffordable . [ccxcvii] 

Right-wing advocates see UBI as a way to remove swathes of government bureaucracy: 
abolishing means testing removes the need for the battalions of civil servants who 
devise and implement it. There would be no incentive for people to game the benefits 
system, thus reducing government-generated waste and unfairness. They hope it would 
facilitate a wholesale simplification of tax structures, and perhaps enable a move to a 
flat tax. And they argue that more lower-income people would go to work because they 
would no longer be caught in benefit traps which penalise them for raising their income 
slightly. This would mean fewer children raised in families where nobody works, a 
particular bugbear of the right, [ccxcviii] 

Most current supporters of UBI are on the left, but it has had support from prominent 
right-wing politicians and economists in the past, notably President Richard Nixon and 
economists Friedrich Hayek and Milton Friedman. 


There have been a surprising number of experiments with UBI: the Basic Income page 
on Reddit lists 25„ [ccxcix] and gives potted descriptions of the purpose and outcomes of 
six of them. [ccc] All the researchers involved reported excellent results, with the 
subjects experiencing healthier, happier lives, and not collapsing into lazy lifestyles or 
squandering the money on alcohol or other drugs. Given that, it is curious that none of 
the experiments have been extended or made permanent. 

The declared purpose of many UBI experiments is to investigate the concern that when 
people receive money for nothing, they stop working. One of the biggest experiments 
conducted so far, involving all 10,000 people in the small town of Dauphin in 
Manitoba, Canada, found that the only two social groups which did stop working were 
teenagers and young mothers, and this was seen as a positive outcome. [ccci] 

Of course, people handing in their notice will be of no concern when machines have 
stolen all our jobs, but a more subtle version of the concern remains: do people in 
receipt of money for nothing stop doing anything of value? Do they become indolent 
couch potatoes, watching TV all day long, or collapse into reliance on alcohol and other 
drugs? Bearing in mind the distinction we made earlier between jobs and work, in a 
world where intelligent machines have automated most economic activity, the question 
is not, do people give up jobs, but do they give up work? 

Unfortunately, none of the UBI experiments carried out so far constitute a rigorous test. 

A rigorous test would be universal, randomised, long-term, and basic - in the sense that 
the income distributed should be enough to live on. [cccii] And so more tests are 

In fact, a number of significant UBI experiments are planned or under way at the time of 
writing. One, in Finland, caused great excitement when it was announced in 2015, but it 
remains unclear how broad-based the experiment will be, and what level of income 
will be paid. The aims are clear, however, and they relate to the right-wing concerns 
listed above. The Finnish researcher in charge of designing it, Olli Kangas, explains 
that the UBI experiment is hoping to demonstrate solutions for three problems with the 
current Finnish benefit system First, people working part-time (perhaps in the gig 
economy) receive neither work-based benefits nor unemployment benefits. Second, 
some people are caught in a benefits trap whereby as their income increases their 
benefits decrease, which removes their incentive to work more and contribute more to 

the economy. Third, the existing benefits system is expensive, requiring too many 
bureaucrats to administer it. 

The sample of Finns who are chosen to receive the UBI will be compared with a 
control sample who are not. Kangas will be exploring their propensity to continue 
working, their reported happiness and well-being, and any changes in their use of health 
and social services. He hopes to recruit a substantial sample - perhaps 100,000, which 
will enable him to detect variations between people of different ages, locations, 
demographics, and employment histories. [ccciii] 

Another interesting UBI experiment is a crowd-funded initiative in Germany, which was 
launched by Berlin-based entrepreneur Michael Bohnmeyer in 2014. [ccciv] By 
December 2015, 26 people had been selected by lottery to receive €1,000 (around 
$ 1,000) a month, paid for by public donations. Most of the recipients reported that it 
didn't change their lives enormously, but they felt less stressed, and in many cases were 
able to embark on creative projects. 

There is no shortage of places keen to experiment with UBI. The Dutch cities of 
Utrecht, Groningen, Wageningen and Tilburg are asking their national government for 
permission to carry out trials, and a referendum is expected during 2016 in 
Switzerland. All these initiatives are looking for ways to tackle problems with existing 
social welfare systems. 

We have to go to Silicon Valley to find an experiment specifically designed to explore 
the impact of UBI in the context of a jobless future when machine intelligence has 
automated most of what we currently do for a living. Just such an experiment was 
announced in January 2016 by Sam Altman, president of the seed capital firm Y 
Combinator, which gave a start in life to Reddit, AirBnB and DropBox. Altman's task 
is not trivial: he will have to figure out a way to quantify the satisfaction his guinea pigs 
derive from their UBI, and whether they are doing anything useful with their time. [cccv] 


With all these experiments bubbling up, the concept of UBI has become a favourite 
media topic, but it is controversial. Many opponents - especially in the US - see it as a 
form of socialism, and the US has traditionally harboured a visceral dislike of 
socialism. (The strong performance of Senator Bernie Sanders, a self-proclaimed 
democratic socialist, in the race to become the Democratic Party's candidate in the 2016 
Presidential election is a striking departure from this norm of US politics.) 

This concern is what seemed to leave Martin Ford somewhat dispirited at the end of his 
book “The Rise of the Robots”. As we saw in chapter 3, he fears that “guaranteed 
income is likely to be disparaged as 'socialism'”, and introducing it will be a 
“staggering challenge”. He is not alone: I have heard similar concerns from a number of 
thoughtful American friends. 

I hope and believe that their fears are over-done. America is, of course, huge - more a 
continent than a country - so generalisations about it are dangerous. But I do not 
believe its people are in general un-thinking or malicious. If and when it becomes 
impossible to deny that a majority (or even a large minority) of its citizens will never 
do paid work again, and for no fault of their own, I do not believe that the rest will 
allow them to starve. 

The dramatic recent changes in American attitudes towards homosexuality and drugs 
show how fast opinions there can change, and how far. As recently as 1962, 
homosexual acts were illegal in every US state, and it was only in 2003 that the federal 
Supreme Court decision in the Lawrence v Texas case invalidated the ban in the last 14 
states where it remained unlawful. (Even today, more than a dozen states have yet to 
repeal or amend their own legislation to reflect this ruling, [cccvij f And yet in June 
2015, the federal Supreme Court ruled that bans on same-sex marriage are 
unconstitutional, in the case of Obergefell v Hodges. According to a Wall Street 
Journal poll, public support for gay marriage has doubled in the last decade, standing 
now at 60%. [cccvii] 

Attitudes towards the legalisation of cannabis have also undergone a rapid sea change. 
For years, governments proclaimed a war on drugs, but that policy has clearly failed. 
Billions of dollars have been spent, and countless lives have been lost, but supply has 
not been constrained, much less eliminated. Parts of Mexico and other countries where 
the drugs are grown or routed have become war zones, and hugely powerful criminal 
organisations have been spawned. Attempts to curb demand have also failed, with tens 
of thousands of people being criminalised for an activity that harmed no-one. Drugs are 
dangerous, and their supply should be regulated, but ceding control over that supply to 
criminal gangs has not proved an enlightened policy. Public opinion in America is 
swinging rapidly towards that position. In 1969 only 16% of voters polled by Gallup 
supported legalisation, but now a majority takes that view. [cccviii] Possession of 
cannabis for personal use is now legal in four states, with the federal government 
agreeing not to interfere . [cccix] 

It is not only America which is experiencing revolutionary changes in social attitudes. 

Up until 1997, sex before marriage was illegal in China, condemned as “hooliganism”. 
Nevertheless, a researcher found in 1989 that 15% of citizens had experienced it. The 
percentage had risen above 70% by 2014. Homosexuality was illegal until 2001 and 
gay marriage is still not legal. But in 2011, state-owned media began writing positive 
articles about gay pride marches in Shanghai and elsewhere, [cccx] 

These examples show that entrenched societal opinions can and do change, sometimes 
quickly. If and when machine intelligence renders many of us permanently 
unemployable, it seems reasonable to expect that opposition to some form of universal 
basic income will evaporate. 


Opponents of UBI also worry that it will stoke inflation. Other things being equal, a 
massive injection of money into an economy is liable to raise prices, leading to sudden 
inflation and perhaps even hyper-inflation. But as campaigner Scott Santens points out, 
UBI does not necessarily mean an injection of fresh cash into the economy. It would 
most likely be paid for by increased taxation of the better-off, and by replacing the 
existing benefits system, together with the bureaucracy which implements it. [cccxi] He 
also claims that where basic incomes have been introduced, as in Alaska in 1982 and 
Kuwait in 2011, inflation actually fell. 


A related objection to UBI which may have more substance is that it is unaffordable. 
Some argue that it can be funded by raising taxes on the small minority who have 
become extraordinarily wealthy in recent years. After all, even some of those wealthy 
people themselves (like Bill Gates and Warren Buffet) have confessed to feeling under- 
taxed. But experience shows that this can be a losing game. Very wealthy people do 
sometimes decide to dedicate much of their wealth to charitable causes. Bill Gates 
(again) and Mark Zuckerberg are obvious examples, and even some of the robber 
barons of the late 19 th century gave fabulous sums to charitable foundations. One of the 
most successful of those barons was the Scottish-American steel magnate Andrew 
Carnegie, who endowed some 3,000 municipal libraries, and provided funding for 
several universities and numerous other organisations before he died. His most famous 
motto was that “the man who dies rich dies disgraced.” [cccxii] But these people 
generally want to determine for themselves how their wealth is deployed, not least 
because they believe that they will make better use of it than politicians and bureaucrats. 

So even the most generously disposed wealthy people often resist the wholesale 
appropriation of their assets in the form of taxation. And as demonstrated by the 
Panama papers scandal that erupted in April 2016, they are well equipped to do so, 
either by hiring clever lawyers and accountants to find loopholes and dodges, or by 
shifting themselves and their assets to less demanding jurisdictions. 

Furthermore, entrepreneurs and other capable commercial people who are not yet 
extremely wealthy but aspire to become so may decide to move out of a jurisdiction 
which raises taxes sharply to pay for UBI. Or if they stay, they may become 
discouraged and decide against taking the necessary risks and dedicating the necessary 
time and energy to projects which could achieve their ambition. These people are 
responsible for much of the dynamism in capitalist countries, and dampening their 
enthusiasm or incentivising them to move elsewhere can be very damaging to an 

This sounds like common sense, but is in fact highly contentious. The political left 
believes that inequality is a social evil, and argues that taxing the rich does not deter 
economic activity, [cccxiii] The political right believes that a modicum of inequality is 
no bad thing, and is anyway inevitable in a thriving economy. It argues that increasing 
taxes on the rich does deter economic activity, and may actually result in lower 
government revenues, as the rich look harder for ways to reduce their tax 
burden, [cccxiv] 

The Laffer Curve 

Unfortunately, the data is muddy, which enables both sides to marshal apparently 
convincing arguments. And as is so often the case, the truth lies somewhere between 
them. We do know that there is a level of taxation beyond which further increases are 
ineffective, or even self-defeating. The Laffer Curve plots tax rates against the revenue 
they raise. At 100%, no-one would work, so that is an inefficient rate; 99% would not 
be much better. Sadly we just don’t know for sure what the optimal level is, either in 
general, or in a specific country at a specific time. [cccxv] 

In the UK, the Labour government in 2010 introduced a top rate of tax of 50% for 
people earning above £150,000. The Conservative government took it down to 45% in 
2013, and claimed the result was a sharp revenue increase. The Labour party, of 
course, claimed the opposite. [cccxvi] 

Which side you choose in this debate will be determined largely by your political 
orientation. Personally, I believe that competing organisations in well-regulated 

markets are more efficient and effective than monopolistic governments, and I believe 
that lower tax regimes encourage entrepreneurship. I also think that governments tend to 
tax their subjects as much as they think they can get away with, which explains why so 
much of their tax take is achieved through subtle, indirect, and often downright stealthy 
taxes. Thus a substantial tax increase to fund UBI is likely to be economically 

Fortunately, this debate is rather tangential to my main argument about UBI, which is 
that we will need to implement it if and when a large minority of people have become 
unemployable. So if you are on the political left, I don’t need to lure you across the 
parliamentary floor, which is probably a relief to us both. 

Before we leave the question of UBI’s affordability, we should consider the claim that 
it can be funded by abolishing existing social benefit arrangements. 

Let’s kill all the bureaucrats 

Channelling Shakespeare, [cccxvii] UBI advocates claim that the massive cost of UBI 
could be offset by abolishing much or all of the existing benefits systems, along with the 
legions of bureaucrats who implement them. They offer an enticing vision of a world 
without means-testing, with no poverty traps, no steely-eyed “advisers” in job centres 
forcing claimants to apply for unsuitable work, no benefit fraud and no need to game the 

Unfortunately the world probably won’t allow such a nice, tidy outcome. People’s 
needs vary according to their capabilities, their life stage, and their location, among 
other things. Someone who is disabled might well suffer greatly if their income was 
equal to that of an able-bodied person in robust health. A single dad with a child may 
need extra support. People living in London or San Francisco would certainly need 
more housing benefit than people living in Albuquerque or Auchtermuchty. Having 
ushered all the bureaucrats out the door thanks to the purifying simplicity of UBI, we 
would have to apologise and call them right back in again. 

The RSA, a British think tank, published a report about UBI in December 2015 which 
was the result of a year’s research and discussions, [cccxviii] It proposed abolishing 
much of the UK’s existing benefits system, and replacing it with a payment of £3,692 for 
everyone between 25 and 65. This is £307 a month, £71 a week, or £10 a day. The 
payment amounts to a modest 14% of the average UK wage, which was £26,500 in 
2015. [cccxix] People aged between 5 and 25 would receive £2,925, and pensioners 

would get £7,420. Extra payments would be made for young children. 

The RSA estimated the total cost of its proposed system at £280bn, including running 
costs of £3bn. It claimed that this would be offset by £272bn saved by abolishing most 
of the existing benefits and pensions infrastructure, including personal income tax 
allowances and tax relief on pensions payments for higher rate tax payers. 

The RSA claimed that families with children and on low wages would be £2,000 to 
£8,000 better off per year because of the removal of benefit traps. Adjustments 
required to prevent poorer people being worse off would take the cost to between £10- 
16bn, around 1% of GDP. This would be funded by taxes on high earners, a group 
which would also lose income from the changes. 

Arguably, the RSA scheme is not a fully-fledged UBI proposal, as payments would 
taper off for incomes above £75,000, and stop altogether at £100,000. The level of 
payments are also set at a level which would keep people alive, but would not provide 
a decent standard of living. 

The system is perhaps more an attempt to simplify and streamline the UK’s messy and 
byzantine benefits system. It is also significant that the proposal ignores payments for 
housing and disability, which are of course substantial, and would require the recall of 
at least some of those bureaucrats. 

Countries are not isolated economic ecosystems. Introducing UBI would significantly 
affect the competitive position of a country which introduced it, and would have other 
unintended consequences. A broadly positive article about UBI in the right-of-centre 
Daily Telegraph newspaper speculated that if and when Finland proceeds with its UBI 
experiment, it will be inundated by economic migrants unless it leaves the EU. [cccxx] 


This review of distribution has focused on income as opposed to wealth. Most people 
have little wealth, and are therefore dependent on income. A poll published in January 
2015 by a US personal finance website [cccxxi] echoed the finding a year earlier by the 
Federal Reserve [cccxxii] that two-thirds of Americans had savings equal to less than 
three months income. Half them could not cover an emergency expense of $400 without 
going into debt. This was aggravated by the recession which began in 2007: the 
average American family’s net worth fell from $136,000 in 2007 to $81,000 in 2013. 

Wealth inequality is far more extreme in today’s world than income inequality, both 

globally and within individual nations. It is also less significant. The charity Oxfam 
created a stir in January 2016 by claiming that the richest 62 people own as much as the 
poorest 50% of the world, [cccxxiii] The figure may or may not be correct, but it tells 
us less than it appears to. A young professional in New York living a life of luxury and 
excess may have no net assets, but it would be perverse to describe her as poor. 
Furthermore, if the richest billionaires gave their wealth to the poorest half of the 
world, it would amount to a one-off payment of few hundred dollars each. [cccxxiv] 

Nevertheless, if you are one of the lucky minority with substantial net assets, you might 
be wondering how you will be affected if and when technological unemployment takes 
hold. Will your house be worth more or less in the new economy? How about your 
vintage Aston Martin, or your collection of fine wines? Until and unless we move to a 
completely different kind of economy, it is likely that some of the wealthy people - 
especially those who control the artificial intelligence which creates most of the added 
value - will remain wealthy, and perhaps become even more wealthy. Perhaps the 
prices for Stradivarius violins and prime real estate will continue to rise - for some 
time at least. 

What about the holdings of the much larger number of people in the middle - people 
who have net assets of a few tens or hundreds of thousands of dollars, perhaps up to as 
much as a million or two? Unless we switch quickly and smoothly to UBI, it seems 
likely that the price of assets typically owned by these middle class people, such as 
suburban houses and mass-produced cars, will slide as their owners try to replace lost 
income by liquidating their property. This could happen quickly, as people look ahead, 
see what is coming, and decide to cash in before the slide starts in earnest. Asset prices 
are notoriously hard to predict because they depend on events which cannot be foreseen, 
and also upon perceptions about what may happen, and perceptions about those 
perceptions. This is another good reason why we should be thinking seriously about 
these matters sooner rather than later, [cccxxv] 

Summary: UBI, but not yet 

My conclusion about UBI will probably be unpopular with many, especially those on 
the political left. It is that the time is not yet ready for a full-fledged UBI, in the sense of 
a payment made to all citizens with no questions asked, and at a level which affords an 
acceptable standard of living in the context of the jurisdiction. UBI is a system which 
requires an economy of abundance, not an economy of scarcity, which is what we still 
have today. The appropriate system for economies of scarcity is the market (with 
regulations), because it allocates resources according to what people actually want, not 

according to what a politician or technocrat thinks they ought to want. 

Any government tempted to experiment with a strong version of UBI must carefully 
consider the possible damage to its country’s economic competitiveness, and other 
international impacts. 

FT journalist Tim Harford summed it up well in May 2016, saying that in current 
circumstances, UBI appeals to three kinds of people: those happy to see the needy 
receive less income, those happy to see the state balloon (and risk massive capital 
flight), and those who can’t add up. [cccxxvi] 

But if and when machines have permanently automated most jobs, we will need to 
implement some form of UBI. The fears about it being politically unacceptable will 
probably prove exaggerated, with attitudes changing as the circumstances change. 

That being so, if and when it becomes clear that most people are going to be rendered 
unemployable, it could be helpful to implement a modest form of UBI - something like 
the RSA scheme mentioned above, perhaps - in order to be able to move quickly to a 
full-blooded version when automation bites deep. 

But UBI will not alone be sufficient to enable us to cope with the end of jobs. The other 
big problem we will have to tackle is cohesion. We will address that later on in this 
chapter, but first we should review the alleged problem of how people find meaning in 
a world without jobs. 

5.3 - Meaning 

The meaning of life 

... is 42, of course, [cccxxvii] 

OK, now we’ve got that out of the way, would you agree with the statement that 
people’s lives need to have meaning in order for them to feel fulfilled, satisfied, and 
happy? It’s certainly true for me, and I’m pretty sure it’s true for most of the people I 
know. It is probably also true of you, or you wouldn’t be reading this book. 

I have met people who claimed to be pure hedonists - interested only in immediate 
pleasure. Some of them may even have been telling the truth. But most of us get bored 
if we feel our lives have no meaning. And not just bored in the sense that you get bored 
in a queue at a supermarket checkout, but profoundly restless and frustrated. To avoid 
this feeling we make deep emotional investments in ideas and institutions like family, 
friendships, work, loyalty to tribes, nations and causes. Deprived of these things, we 
feel lost and alienated. 

Perhaps the most famous quote attributed to the 4 th century BC Greek philosopher 
Socrates is that the unexamined life is not worth living. It is a remarkably strong 
statement. Why not just say that an unexamined life - a life without philosophy, in other 
words - is less good than an examined one? Is an unexamined life really worse than 
death? He made the statement at his trial, when the outcome was a choice between exile 
and suicide (he chose suicide), so perhaps he was under stress and being hyperbolic. 

But the claim is usually taken at face value, and perhaps he meant it literally. 

It is also an elitist statement. Many people are too preoccupied with making a living, 
raising a family, escaping drug addiction or whatever immediate challenge they face to 
indulge in the luxury of philosophical discourse. Are their lives not worth living? You 
could argue that Socrates and his fellow ancient Athenians had slaves to take care of the 
menial stuff, but we have labour-saving devices instead, so that’s no excuse. 

Of course the question of what constitutes a good life, a worthwhile life, a life with 
meaning is a vexed one, with no simple answers, and probably no single answer. The 
philosopher John Danaher distinguishes between subjective accounts, which involve 
feeling worthwhile, and objective accounts, which involve helping to make or do 
something worthwhile, [cccxxviii] 

Despite not knowing (or at least not agreeing) what a meaningful life is, and despite not 
spending all that much time in the average day thinking about it, most of us believe we 
need it. And many of us find it in work. So it’s going to be a problem if we stop 

Or is it? 

Meaning and work 

Simon Sinek has made a name for himself with books that propound a simple but 
important truth: if you have a clear purpose which inspires others, you can achieve great 
things. His best-known saying is “Working hard for something we don't care about is 
called stress; working hard for something we love is called passion.” 

You could be forgiven for thinking that a law was passed a few years ago in the US 
requiring business leaders - and people who want to be business leaders - to talk about 
their passion for their business. But most people don’t feel passionate about their work, 
even if they pretend they do. In fact, many people are positively alienated by their jobs. 
They find them meaningless and boring. 

Yet even these people usually define themselves by what they do for a living. If you ask 
someone at a party what they do they are likely to reply that they are an accountant, a 
taxi driver or an electrician. They are less likely to say that they are the coach of their 
child’s football club, or a cinema-goer, or a reader. No doubt this is partly due to the 
amount of time that our jobs absorb - but then again we don’t define ourselves as 
sleepers. It also has to do with work being the activity that provides our income, which 
is why home-based parents often feel sheepish about naming that as their work. (They 
shouldn’t: it is some of the hardest but most rewarding work I’ve ever done!) 

So work helps define us, and it gives many of us purpose. It even gives some of us 
meaning. So how damaging would it be if we lost it? Unemployed people often 
struggle with depression, but they are experiencing it in the context of a society where it 
seems that everyone else has a job. They are also on a lower income than the employed 
people around them How bad would it be if everyone else was also unemployed, and 
receiving a decent income? 

Fortunately, there are a couple of places we can look for an answer to that question. 

The rich and the old 

The agricultural revolution, around 12,000 years ago, created sustainable surpluses of 
food and other basic resources. This enabled a class of people to stop doing the work 
that pretty much all humans had done since our arrival on the planet, which was foraging 
and hunting for food. They became tribal leaders, kings, warriors, priests, traders and 
so on. Sometimes they spent as much time on these activities as the people who 
continued to forage and hunt, but sometimes they took time off- deliberately or by 
happenstance - and engaged in lives of leisure. 

In Europe these people became known as aristocrats, from a Greek word meaning the 
best - originally in a military sense and then a political one. Some aristocrats did jobs: 
they ran agricultural concerns, they got involved in politics, and in some countries they 
ran empires. Occasionally they became men (and more rarely, women) of science. 
Famously, they disdained trade and commerce, regarding those activities as the 
preserve of the class below them, the middle class. 

Many aristocrats did not work - including almost all the female ones. They led lives of 
leisure. As young men (and in a few cases, young women) they toured classical Greek 
and Roman sites in the Mediterranean countries. Returning home, they mostly 
socialised. Their lives revolved around balls, hunts, and visits to their local peers, 
interspersed with the glamour and tragedy of war, if that was their inclination. This 
lifestyle was chronicled in the novel, an art form which first acquired its current 
realistic form in the early 18 th century, [cccxxix] 

The lives depicted by Jane Austen and her contemporaries may seem tame to modern 
readers, who have experienced international travel and expect simultaneous global 
communications. But they were agreeable lives compared to what their poorer 
contemporaries had to put up with. Addictions to gambling and drink were a hazard, 
and of course a minority of this pampered class destroyed themselves and their families 
with these vices. But this was unusual, and by and large most 18 th - and 19 th -century 
European aristocrats seem to have passed their lives without great concern about their 
lack of meaning. Whether these lives were worthwhile or not, whether or not they had 
meaning, is probably not for us to judge, but there is no evidence of widespread 
existential angst among the nobility. 

In fact, it is these privileged people who made most of the advances in human thought 
and art in previous centuries, precisely because they did not need to work for a living, 
or eke out an existence as subsistence farmers. If they did not produce the memorable 

work themselves, they often sponsored it by employing talented artisans. So it seems 
there is much to be said for the ability to be idle. [cccxxx] 

The other group we can look to for evidence about the effects of joblessness are retired 
people. The conventional wisdom used to be that growing old was an almost 
unmitigated disaster: “Old age ain’t no place for sissies”, as Bette Davies 
said. [cccxxxi] although it’s obviously better than the only alternative currently 
available. But starting in the 1990s, researchers began questioning this perception, and 
found instead that the progress of happiness throughout life is U-shaped. We are at our 
happiest and most fulfilled when young, we become stressed and discontented in our 
prime and middle age, and we are happier and more relaxed again when older, despite 
the onset of physical disabilities and limitations . [cccxxxii] This pattern has been 
observed across a wide range of societies, and over a substantial period of time. 

There are probably numerous causes of this effect, including the relinquishment of 
responsibility for children, and the acquisition of wisdom to accept what life has thrown 
at us. But the absence of jobs plays a major role in the lives of the retired. Even if it is 
not causing the up-tick in happiness, it is at least not preventing it. 

Virtually happy 

Thus far in human history we have had to find our meaning within the constraints of the 
three-dimensional world we live in, or in our imaginations. Technology is poised to 
open up a whole new space for us to explore together - the world of virtual reality. We 
don’t yet know how we will react to this new universe, how we will behave in it, and 
what it will mean to us. We can be pretty confident that it will have a big impact. 

“Diaspora”, Greg Egan’s novel of the far future, features an environment called the 
Truth Mines. It is a physical representation of mathematical theorems (albeit in virtual 
reality) which can seemingly be explored forever without exhausting all the discoveries 
that can be made. The ability to create virtual worlds that are so convincing to our 
brains that we almost lose the understanding that they are artificial may well allow us to 
expand enormously the space within which we find happiness and meaning. 

In summary, loss of meaning does not seem likely to be one of the biggest problems that 
widespread technological unemployment will create. 

5.4 - Allocation 

The house on the beach 

In a world where the majority of people cannot get jobs and are therefore paid a 
universal basic income, how will we allocate goods and services? At first sight, the 
answer seems simple and obvious: we will still have money, so we will still have the 
market. Supply and demand will continue to operate like before. 

Let’s assume we’re at the point where a large majority of people get all their income 
from UBI. There would have to be adjustments for people looking after children or 
other dependents, disability and so on, but any extra income allocated for these needs 
would also be swallowed up by those needs. So in income terms, the society we are 
discussing would be an extremely egalitarian one. 

But this society of egalitarian incomes will have inherited a decidedly un-egalitarian 
asset base. Houses are the most obvious and the most significant example: they were 
not all created equal. In a society where everyone’s income is more-or-less the same, 
how will we decide who lives in the nice big house in the posh part of town, and who 
lives in the small flat with no sound-proofing in the grubby apartment block in the 
unfashionable suburb? 

Will it be like a game of musical chairs? We all work hard to improve our lot, and then 
when the machines take our jobs, the music stops and we all sit down in the chairs we 
have arrived at. And just stay there forever. That seems neither fair nor sustainable. 

With luck, we will be creating an economy of abundance, in which machines carry out 
maintenance and improvement works with great efficiency, and hence cheaply. In that 
case we will set them to revitalising and / or replacing the stock of lower-quality 
houses. (And cars, and boats, and furniture, and clothes, etc.) But it will take a very 
long time indeed to build a nice new house for everyone who doesn’t start off with one. 
And even when we have completed that gargantuan task, some houses will still be in 
much nicer places than others. 

This is the allocation problem. 

Some scarcity can never be abolished. There is a finite and regrettably small supply of 
large houses on empty white sand beaches fringed with palm trees leading down to a 
turquoise sea. Or penthouse apartments on Manhattan’s Fifth Avenue. There is a very 

small supply of Vermeers and Aston Martin DB5s. Do we decide that no-one can own 
these things? Perhaps we could turn all the nice houses into museums and keep the 
scarce movable objects on display there, to be visited (and perhaps used) on payment of 
a fee, or by scheduled appointment. In that case, who will decide what the cut-off point 
is between a house which people can carry on living in, and one which is too nice to be 
private property? 

VR to the rescue? 

At the time of writing, Palmer Luckey and John Carmack are hardly household names, 
but by the time this book is published they may well be. (In case they’re not, they are 
the key executives of Oculus Rift, which looks set to be the first commercially-available 
VR equipment to offer a convincingly immersive user experience.) They talk about a 
“moral imperative” to make virtual reality available to us all. [cccxxxiii] 

Luckey puts it like this: “Everyone wants to have a happy life, but it’s going to be 
impossible to give everyone everything they want.... Virtual reality can make it so 
anyone, anywhere can have these experiences.” Carmack continues: “you could imagine 
almost everyone in the world owning [good VR equipment]. ... This means that some 
fraction of the desirable experiences of the wealthy can be synthesized and replicated 
for a much broader range of people.” 

Other people have thought about these questions, and not everyone is delighted by the 
suggestion that VR can assuage the frustration caused by scarcity. Some people think it 
impossible, and others think it possible but degrading. 

The Harvard political philosopher Robert Nozick described a thought experiment back 
in 1974 featuring an “experience machine” which could recreate any sensation you 
choose. Your brain is persuaded that the experience is real, which means that you 
believe it too, but in fact your body is lying in in a flotation tank, deprived of all sensory 
input while your brain is hooked up to the machine. Philosophers do a lot of their work 
by investigating their intuitions, and Nozick’s intuition was that no-one would use this 
machine because we value reality too highly. I find it surprising that he came to that 
conclusion back in 1974, and it would be an even more surprising conclusion to reach 
today, when so many people spend so much of their lives in simulated realities, albeit 
only imperfectly simulated. Certainly a great deal of money is being invested by smart 
people in the belief that we will consume VR avidly. Nozick died in 2002, so he won’t 
have to find out for himself- maybe he would be relieved. 

Other critics see the Oculus founders’ view of the future as possible but frightening. 
Ethan Zuckerman is director of the MIT Centre for Civic Media, and thinks that “the 
idea that we can make gross economic inequalities less relevant by giving [poor 
people] virtual bread and circuses is diabolical and delusional.” Jaron Lanier is a 
computer scientist and writer who founded VR pioneer VPL Research, and is generally 
credited with popularising the term virtual reality. He lambasts as “evil” the vision that 
the rich will become immortal, while “everyone else will get a simulated reality. ... 

I’d prefer to see a world where everyone is a first-class citizen and we don’t have 
people living in the Matrix.” 

Only time will tell if VR is helpful, or even necessary, in enabling us to live in a world 
where machines have made humans unemployable. My own guess is that it will play a 
major role in the lives of most people, and that it will make them more productive, more 
fun and more fulfilling. As Oculus’ John Carmack puts it, “if people are having a 
virtually happy life, they are having a happy life. Period.” 

Nevertheless, Zuckerman and Lanier have identified an important problem with the 
vision. It is not to do with VR so much as the potential separation of our species into 
two or more divergent camps. We will review this in more detail in the next section. 


Decisions about the allocation of resources are being made all the time in societies, on 
scales both large and small. As argued above, capitalism has proved so effective in 
raising the living standards of societies which have adopted it (paraphrasing Churchill 
agai n[cccxxxiv] . it is the worst possible economic system except for all the others) 
because markets are highly efficient systems for allocating resources in economies 
characterised by scarcity. 

Historically, markets have consisted of people. There may be lots of people on both 
sides of the transaction (flea markets are one example, eBay is another). Or there may 
be few buyers and many sellers (farmers selling to supermarket chains) or vice versa 
(supermarket chains selling to consumers). But typically, both buyers and sellers were 
humans. That is changing. 

Algorithms, as we saw in chapter 3, are sets of rules or instructions for a computer to 
follow. A model is built and refined by testing with data. (Lots of data!) In situations 
with well-defined processes and target outcomes, algorithms can be extremely efficient 
decision-makers. We see this in our everyday lives. ATMs are operated by algorithms 

- no human responds to the instruction you type into the screen. They are efficient and 
effective, and most of us prefer to draw our cash from ATMs rather than queueing at a 
counter to have a human provide the service. 

Algorithms now take many decisions which were formerly the responsibility of 
humans. They initiate and execute many of the trades on stock and commodity 
exchanges. They manage resources within organisations providing utilities like 
electricity, gas and water. They govern important parts of the supply chains which put 
food on supermarket shelves. 

With the rapidly increasing volume of data flowing from tiny sensors embedded in 
buildings, machinery, vehicles and all kinds of products, and with the ever-improving 
performance of machine learning systems thanks to Moore’s Law, algorithms are getting 
better and better at their jobs. They are getting better, cheaper and faster than humans. 
As well as taking our jobs they are taking our decisions. 

In his 2006 book “Virtual Migration”, Indian-American academic A. Aneesh coined the 
name “algocracy” for this phenomenon, [cccxxxv] The difficulty with it has been 
explored in detail by the philosopher John Danaher, who sets the problem up as 
follows. Legitimate governance requires transparent decision-making processes which 
allow for involvement by the people affected. Algorithms are often not transparent and 
their decision-making processes do not admit human participation. Therefore 
algorithmic decision-making should be resisted, [cccxxxvi] 

Danaher thinks that algocracy poses a threat to democratic legitimacy, but does not think 
that it can be, or should be, resisted. He thinks there will be important costs to 
embracing algocracy and we need to decide whether we are comfortable with those 

Of course many of the decisions being delegated to algorithms are ones we would not 
want returned to human hands - partly because the machines make the decisions so much 
better, and partly because the intellectual activity involved is deathly boring. It is not 
particularly ennobling to be responsible for the decision whether to switch a city’s 
street lights on at 6.20 or 6.30, but the decision could have a significant inpact. The 
additional energy cost may or may not be offset by the improvement in road safety, and 
determining that equation could involve collating and analysing millions of data points. 
Much better work for a machine than a man, surely. 

There are many decisions which machines could also make better than humans, but we 
might feel less comfortable having them do so. The allocation of new housing stock, the 

best date for an important election, the cost ceiling for a powerful new drug, for 
instance. Arguments are probably going to become increasingly commonly and 
increasingly vehement over which decisions should be made by machines, and which by 

5.5 - Cohesion 

The scenario of “the Gods and the Useless” 

As mentioned in chapter 1, at the end of his July 2015 TED talk. [cccxxxvii] the author 
of “Sapiens”, Yuval Harari, makes a seemingly throw-away comment about humanity 
devolving into two classes: the gods and the useless. The audience laughs at this brutal 
assessment, but I suspect Harari is deadly serious. 

Imagine a society where the great majority of people lead lives of leisure, their income 
provided by a beneficent state, or perhaps a gigantic charitable organisation. They are 
not rich, they don’t travel first class or frequent expensive restaurants, and they don’t 
own multiple houses. But they have no pressing needs and in fact they want for little: 
they enjoy socialising, learning, sports, exploration, and much of this is carried out in 
virtual worlds which are almost indistinguishable from reality. 

A small minority of people in this society do have jobs. Their work is pleasurable and 
intellectually simulating, and not stressful. It involves monitoring and occasionally 
guiding or re-setting the performance of the machines which run their society - machines 
which they own. 

Let’s say that this elite minority is generous towards the majority which lives outside 
their gated communities, and which does not visit the luxurious resorts they migrate 
between, and does not travel with them on their private heli-jets. They are effectively 
benign rulers, although both camps refrain from putting it like that. 

Now, in this future world, all members of the species of homo sapiens are changing. 
They are using new technologies to enhance themselves both cognitively and 
physically. They use smart drugs, exoskeletons and genetic technologies, among others. 
Maybe they have engineered themselves to need less sleep . [cccxxxviii] 

Everyone has access to these technologies, but the elite has privileged access. They get 
them sooner, and this could be vitally important. I argued in chapter 3.7 that concerns 
about a “digital divide” were exaggerated. Companies make much more money by 
selling lots of relatively cheap cars and smartphones to almost everyone than they ever 
could by selling just a handful of diamond-encrusted versions to the super-rich. 

But we mustn’t forget that technology is advancing at an accelerating rate. In the future 

society we are envisioning, important breakthroughs in physical and cognitive 
enhancement are announced every year, then every month, then every week. As 
artificial intelligence gets better and better it fuels this improvement - even though it 
(the AI) is still narrow AI, and far from becoming human-level, artificial general 
intelligence, or AGI. [cccxxxix] 

It may become hard or even impossible to disseminate these cognitive and physical 
improvements quickly enough to avoid a profound separation between those with 
privileged access to them and the rest of us. 

So the elite will change faster than the rest. As the two groups lead largely separate 
lives, the widening gap may not be apparent to the majority, but the elite surely will 
know about it. They will decide that they must draw attention away from the fact, and 
they will take precautions to prevent attack, in case the majority should become aware 
of and resentful about what is happening. They will surround their gated communities 
with discreet machines which possess astonishingly powerful defensive - and offensive 
- capabilities. They will keep themselves more and more to themselves, meeting 
members of the majority less and less often. When they do meet, it will almost always 
be in virtual reality, where their avatars (their representations in the VR environment) 
do not betray the widening gulf between the two types of humans. 

After a while, humans will evolve into two different species: the gods and the useless. 

Brave New World 

This is not your average science fiction dystopia. The scenario most commonly offered 
up by Hollywood is that technology has pulled back the curtain which concealed the 
truly evil nature of the capitalism system, and mankind has fallen into a form of high- 
tech slavery, where the rich descendants of company CEOs and scheming tycoons and 
financiers brutalise an impoverished and oppressed majority. The movie “Elysium” is 
just one of many dreary examples of this tired old cliche. In it, as in so many others, 
society has actually regressed from capitalism into a sort of techno-feudalism. Any 
viewer who is half-awake is wondering, since machines can do all the work, what is the 
point of enslaving humans? 

(It is curious that the left-leaning culture prevalent in Hollywood impels it to issue these 
tirades against capitalism, when Hollywood studios are themselves formidable 
exponents of the capitalist art. And it is curious that Hollywood stars who demand 
millions of dollars to make a single film complain that corporations are fuelled by 

nothing but greed.) 

A far more interesting scenario is Aldous Huxley’s “Brave New World”, which is a 
very subtle piece of world-building. Almost everyone is content in the society which 
has been developed after an appalling military conflict, but it is clear to the reader that 
humanity has lost something important, and most of it has regressed to an almost 
infantile condition. Yet when a talented outsider arrives, he is unable to devise a way 
to improve the system, or to accommodate himself to it. He instinctively feels - and the 
reader is encouraged to agree - that his own life has shown him there is a better way to 
live, but he is unable to articulate or maintain it. 

In Huxley’s story, humanity has achieved a stable equilibrium. Only a tiny minority of 
humans are aware of what has been sacrificed in order to achieve a society of such 
docility and acquiescence. Regular sex and a powerful drug called soma are the opiates 
of the masses. (Huxley wrote the book in 1931, before the advent of rock and roll, so he 
couldn’t anticipate Ian Dury’s hedonistic mantra that sex and drugs and rock and roll 
were all his brain and body needed, [cccxl] ) 

Brave New World is certainly not intended as a blueprint. Even so, it is calm, and the 
spectre of economic and social collapse seems to have been abolished - at least for the 
time being. It would be foolish of us to take even that much for granted. A society 
comprising gods and the useless might turn out to be inherently unstable. In a full-on 
conflict between them it seems likely that the gods would have the means to protect 
themselves, but at what cost? 

Will capitalism remain fit for purpose? 

Private property is an essential feature of capitalism, and in particular, the private 
ownership of the means of production, exchange and distribution. In market economies, 
most people earn their living by selling their labour - their time and their physical and 
intellectual skills. 

People called entrepreneurs hire workers and combine their labour with the other major 
element of the capitalist economy - capital, which consists of money, machinery, land, 
buildings and intellectual property. The entrepreneurs use the labour and the capital to 
develop and make products and services which they hope will be bought by enough 
people to turn a profit. The capital is put at risk during this process, and the profit is a 
reward and an incentive for the entrepreneurs and the owners of the capital (known as 
capitalists) who took that risk. 

The picture is complicated, because we are all capitalists today. Much of the capital 
deployed by entrepreneurs is owned by financial institutions like insurers and pension 
companies, and ownership of these institutions is very widely distributed through 
pension plans. Many people are also shareholders in the companies they work for, 
thanks to employee share ownership schemes. 

In a capitalist economy where most people work, individuals who start off without any 
capital can acquire it by saving some of what they are paid for their labour, and by 
starting companies themselves. Countries where this is easy and celebrated tend to be 
better off than countries where entrepreneurship is discouraged, or hampered by 
corruption, over-regulation, or lack of infrastructure. 

Similarly, people who start off with capital can lose it, by bad luck or poor judgement. 

It is not uncommon for families to go from rags to riches and back to rags within three 

There are countries today where this kind of social mobility (both upward and 
downward) is very limited, and the elites are entrenched. Arguably, nowhere in the 
world has sufficient economic social mobility, but in many countries it does exist to 
some degree. 

As we have seen, after the economic singularity there will be two major differences. 
First, if the prediction that most people will be unemployable comes true, then there 
will be pretty much no traffic - no economic and social mobility - between the elite and 
the rest. This will be the case all over the world. If you can’t do paid work, it is very 
hard to accumulate capital. Second, if the rate of technological progress continues to 
accelerate, the elite may avail themselves of the means of cognitive and physical 
enhancement to diverge from the majority, both physically and cognitively. 

The obvious but difficult remedy for this is to end the institution of private property. 

The means of production, exchange and distribution would be placed into some kind of 
collective ownership to prevent the possibility of social and species fracture. 

As we saw in chapter 3.1, this conclusion is rejected by the two most popular books 
published so far about technological unemployment. I share their inclination, and it 
makes me extremely uncomfortable. I was in business for 30 years before becoming a 
full-time writer and speaker, and I remain convinced of the largely positive effects of a 
regulated market economy with a welfare safety net. [cccxli] It seems to me that 
capitalism is the best economic system we know of for a society where humans do the 

But I fear that capitalism may not be so well-suited for an economy of abundance, where 
machines do the work, where most people are unemployed, and where technology is 
changing the species quickly. I am not pushing this argument hard. Not only does it 
make me uncomfortable, but one of the fundamental characteristics of a singularity is 
that it is even harder than usual to predict the future when there is an event horizon in the 

It is by no means certain that abandoning capitalism and private property is the only way 
to avoid fracture and collapse. It may be possible for all kinds of people to live in 
harmony in a society where a minority gets paid to work and owns most of the 
economy’s assets, while everyone else lives happily on a universal basic income. It is 
not impossible to imagine a benign technocracy in which the majority of people really 
don’t care who owns what, because they are wholly satisfied with their abundant supply 
of material and digital goods and services. 

Unfortunately, every time I try to envisage this world, the picture degrades into a 
variation on the theme of “Brave New World” - or worse. Perhaps this is simply a 
failure of my imagination. I hope so. 

If it is true that we need to move away from capitalism, we have two major jobs on our 
hands. First, we need to determine what that new economy should look like. Second, 
we need to work out how to transition from the economy we have to the economy we 
need. This will not be easy. Humans dislike change, and as always, there will be 
winners and losers. The losers may not take their losses calmly. 

The scenario of the gods and the useless is not the only possible outcome of 
technological unemployment. The next chapter explores half a dozen of the most 
plausible scenarios. By assessing the likelihood and utility of each scenario, and 
understanding how to achieve or avoid them, we may be able to achieve the most 
positive outcome of the economic singularity. 

6. - Scenarios 

6.1 - No Change 

In a July 2015 interview with Edge, an online magazine, Pulitzer Prize-winning veteran 
New York Times journalist John Markoff lamented the deceleration of technological 
progress - in fact he claimed that it has come to a halt. [cccxlii] He reported that 
Moore’s Law stopped reducing the price of computer components in 2013, and pointed 
to the disappointing performance of the robots entered into the DARPA Robotics 
Challenge in June 2015 (which we reviewed in chapter 3.7). 

He claimed that there has been no profound technological innovation since the invention 
of the smartphone in 2007, and complained that basic science research has essentially 
died, with no modern equivalent of Xerox’s Palo Alto Research Centre (PARC), which 
was responsible for many of the fundamental features of computers which we take for 
granted today, like graphical user interfaces (GUIs) and indeed the PC. 

Markoff grew up in Silicon Valley and began writing about the internet in the 1970s. 

He fears that the spirit of innovation and enterprise has gone out of the place, and 
bemoans the absence of technologists or entrepreneurs today with the stature of past 
greats like Doug Engelbart (inventor of the computer mouse and much more), Bill Gates 
and Steve Jobs. He argues that today’s entrepreneurs are mere copycats, trying to 
peddle the next “Uber for X”. 

He admits that the pace of technological development might pick up again, perhaps 
thanks to research into meta-materials, whose structure absorbs, bends or enhances 
electromagnetic waves in exotic ways. He is dismissive of artificial intelligence 
because it has not yet produced a conscious mind, but he thinks that augmented reality 
might turn out to be a new platform for innovation, just as the smartphone did a decade 
ago. But in conclusion he believes that “2045... is going to look more like it looks today 
than you think.” 

It is tempting to think that Markoff was to some extent playing to the gallery, wallowing 
self-indulgently in sexagenarian nostalgia about the passing of old glories. His critique 
blithely ignores the arrival of social media and much else, and dismisses the basic 
research that goes on at Google X, DeepMind, the Human Brain Project and elsewhere. 

Nevertheless, Markoff does articulate a fairly widespread point of view. Many people 
believe that the industrial revolution had a far greater impact on everyday life than 
anything produced by the information revolution. Before the arrival of railroads and 
then cars, most people never travelled outside their town or village, much less to a 
foreign country. Before the arrival of electricity and central heating, human activity was 
governed by the sun: even if you were privileged enough to be able to read, it was 
expensive and tedious to do so by candlelight, and everything slowed down during the 
cold of the winter months. 

But it is facile to ignore the revolutions brought about by the information age. 

Television and the internet have shown us how people live all around the world, and 
thanks to Google and Wikipedia, etc., we now have something close to omniscience. 

We have machines which rival us in their ability to read, recognise images, and process 
natural language. And the thing to remember is that the information revolution is very 
young. What is coming will make the industrial revolution, profound as it was, seem 
pale by comparison. 

As we stressed at the start of chapter 4, the future is unknown, and all predictions are 
perilous. But the idea that the world will be largely unchanged three decades hence 
seems the least plausible of the scenarios set out in this chapter. 

6.2 - Racing with the machines 

We came across the idea of centaurs in chapter 3.10. The notion comes from Gary 
Kasparov, the chess grandmaster who was beaten by IBM’s Deep Blue computer in 
1997. He noticed that (for the time being at least) a human teamed up with a machine 
can beat the best chess-playing computer, and he called the combination a centaur. It 
has since become a metaphor for the hope that humans will not be rendered 
economically redundant by intelligent machines, but instead will work ever more 
closely with them, always bringing some special human magic to the combination. 


As we saw in chapter 3.9, this hope that we can be centaurs gains some credence from 
the early experience with AI systems in what lawyers call discovery, or disclosure. 
After a few hours or days of training, deep learning systems are dramatically quicker 
and better than humans at ploughing through huge piles of documents, looking for 
particular pieces of information. It was feared that these systems would remove the 
need for junior lawyers, but instead it turned out that the machines brought a huge new 
class of work into the realm of possibility. Projects that would previously have been 
uneconomic were now feasible, and the junior lawyers are still needed to carry out the 
initial training. This has been called the iceberg phenomenon: it was thought the junior 
lawyers were standing on thin ice, but it turned out instead that they were standing on 
top of a massive bulk of newly available work. Their position began to look secure 

A similar phenomenon is likely to be observed with medical diagnosis. Again in 
chapter 3.9, we saw that cheap attachments to smartphones will soon enable us all to be 
tested far more often and far more cheaply than at the moment. The AI resident on your 
smartphone - or servers it accesses in the cloud - will assess your blood pressure, 
blood glucose, your breath, your voice and many more indicators, and deliver instant 
verdicts in most cases. Initially at least, the upshot will be that we will have much 
better information about our medical condition, and we will still need doctors to carry 
out the more significant tests. 

Creativity and caring 

Sceptics about technological unemployment also argue that human creativity will remain 
in demand, as will human empathy, which they see as a pre-requisite in caring 
professions like nursing. Senior people in the technology companies which are 
developing AI often make this argument. Microsoft's principal researcher Jonathan 
Grudin, for instance, says that “technology will continue to disrupt jobs, but more jobs 
seem likely to be created. There is no shortage of things that need to be done and that 
will not change, " [cccxliii] It is not surprising to hear these arguments from executives 
in businesses which are transforming themselves into AI companies: they would 
presumably feel very uncomfortable if they thought their work was hastening an 
economic crisis. 

But while technology company executives sound breezy about the prospects for 
continued unemployment as machine intelligences get smarter, some of the academic 
authors who broadly agree with them sound more tentative. In chapter 3.1 we saw that 
in their book “The Second Machine Age”, MIT professors Erik Brynjolfsson and 
Andrew McAfee believe that for many years to come, humans will be better than 
machines at generating new ideas, and complex forms of communication. They think 
that capitalism should be defended and retained, but they sound less confident about 
what will happen in the medium term. They argue for an overhaul of the US education 
system, but they don’t sound convinced that will be enough, and they speculate that a 
negative income tax may eventually become necessary. 

Tyler Cowen, whom we encountered in chapter 3.3 as the author of “Average is Over”, 
is certainly not breezy in his assessment of the outlook, nor is he tentative. He is 
confident that UBI will not be needed, and he does not expect riots. But his prognosis is 
lugubrious. He foresees 10-15% of the population being extremely wealthy, and the rest 
getting by on incomes which are stagnant at best, but putting up with it because many of 
them are too old to riot, and they are pacified by the excellent cheap entertainment that 
technology provides. 

We addressed these arguments in chapter 3.10, and we need not go over them in detail 
again now. In summary, the danger is that the icebergs will stop growing after a while, 
and the machines will need less and less training to tackle each one. The 5m Americans 
who currently earn a living by driving are not all going to become computer 
programmers or custodians of the AIs, and it is hard to believe that we can all become 
professional artists, nurses and therapists. 

6.3 - Capitalism + UBI 

Martin Ford, author of not one but two books on technological unemployment, also 
believes that capitalism can survive the “rise of the robots”, as he calls it in the title of 
his second book. He thinks it will be a struggle for many people, though, so he urges 
policy-makers to start thinking about introducing a Universal Basic Income (UBI) of 
around $10,000 a head when the time is right. He believes it will be extremely 
challenging to introduce this policy in the US, given the ingrained hostility to socialism 

But as we saw in the last chapter, there are reasons to suppose that this challenge may 
evaporate if technological unemployment bites severely. And as we also saw in the last 
chapter, it may be replaced by challenges which are much harder to overcome. 

6.4 - Fracture 

This is the “gods and the useless” scenario which we discussed in chapter 5.5 about the 
challenge of cohesion. 

6.5 - Collapse 

Civilisation is fragile. Any schoolchild can name some great empires which collapsed: 
the Romans, the Maya, the Persians. The ancient Egyptians managed to rise and fall 
several times during their extraordinary 3,000-year history. 

We also know how fragile civilisation is from two famous episodes in experimental 
psychology. In 1961, Yale psychologist Stanley Milgram recruited students from that 
elite university, and told them to administer mild electric shocks to incentivise a 
stranger who was supposed to be learning pairs of words. The shocks were fake, but 
the students did not know this, and an extraordinary two-thirds of the students were 
prepared, when urged on by the experimenter, to deliver what appeared to be very 
painful and damaging doses of electricity, [cccxliv] The experiment has been replicated 
numerous times around the world, with similar results. 

Ten years later, Stanford psychology professor Philip Zimbardo, a school friend of 
Milgram’s, ran a different experiment in which students were recruited and arbitrarily 
assigned the roles of prisoners and guards in a make-believe prison. He was shocked to 
see how enthusiastically sadistic the students who were chosen to be guards became, 
and he was obliged to terminate the exercise early, [cccxlv] This experiment has also 
been replicated numerous times. 

Our 21 st century global civilisation seems pretty robust. We have just gone through 
what is frequently described as the worst recession since the Great Depression of the 
1930s, and for the great majority of people, the experience was nothing like as awful as 
those terrible years, which did so much to set up the disastrous carnage of World War 

But history and experimental psychology demonstrate that we cannot afford to be 
complacent. If the argument of this book is correct, we are about to embark on a journey 
towards a new type of economy which we have not yet designed. Unless we are 
careful, there will be plenty of opportunities for mis-steps, misunderstandings, and 
downright mischief by populists and demagogues. 

If technological unemployment arrives in a rush, and we are not prepared, a lot of 
people will lose their incomes quickly, and governments may not move fast enough to 
avert drastic collapses in asset prices as people sell their belongings to make ends 
meet. If the introduction of UBI is slow or botched in some countries, the resulting 

economic crises could lead to their governments being overthrown by irresponsible or 
foolish leaders. We must hope this does not happen in any countries with significant 
stocks of nuclear weapons. 

6.6 - Protopia 

Utopia, Dystopia, Protopia 

Kevin Kelly is a writer, and the founding editor of Wired magazine. He has been called 
the most interesting man in the world. [cccxlvi] I’m not sure whether he enjoys the 
burden of that appellation, but he does produce a lot of interesting ideas. One of his 
good ones is Protopia. 

Too much of today’s thinking about the future is dystopian, and that is partly because too 
many people fail to realise just how much progress homo sapiens has made in the last 
few centuries and decades. It is natural and indeed helpful for our species to be 
discontented: if we weren’t discontented, we probably wouldn’t struggle to make the 
world a better place. But it can lead to misunderstandings. 

Many people think that all politicians are corrupt, and that all corporations are run by 
Bond villains who are greedy and bent on world domination. Most of us could think of 
some group, clique or tribe that we are suspicious, fearful or disdainful of. But the truth 
is that in most of the world, today is the best time there has ever been to be alive. Most 
people in developed countries today live better than kings and queens did a couple of 
centuries ago. We live longer, eat better, have better healthcare, and inconceivably 
better access to information and entertainment than previous generations. Of course 
everything could go horribly wrong tomorrow. There might even be an iron law of 
nature that when civilisations reach a certain stage they either blow themselves up, or 
create machines to do it for them. But from where we stand today, there is no reason to 
believe that. It seems more likely that the future is open, and potentially very good 

Utopian visions of the future are less common, but also problematic. A future in which 
life has to all intents and purposes become perfect sounds sterile and boring. It is also 
highly improbable: the more we learn about the universe, the more we discover that we 
don’t know, so it seems unlikely the universe will one day stop presenting us with 
puzzles and challenges. Perhaps this is why the two best-known literary descriptions of 
utopias, Thomas More’s “Utopia” and Voltaire’s “Candide”, are essentially critiques of 
the societies they lived in rather than recipes for an ideal future one. 

So it is refreshing to read Kelly saying this: “I am a protopian, not a utopian. I believe 
in progress in an incremental way where every year it's better than the year before but 

not by very much—just a micro amount. I don't believe in utopia where there's any kind 
of a world without problems brought on by technology. Every new technology creates 
almost as many problems that it solves.” But crucially, it gives us “a choice that we did 
not have before, and that tips it very, very slightly in the category of the sum of 
good.” [cccxlvii] 

Collective ownership 

What might a society look like which has passed successfully through the event horizon 
of the economic singularity? Machines are producing virtually all the goods and 
services that we would otherwise pay people to produce, and the society has avoided 
the twin traps of fracture and collapse. Everyone shares in the bounty that the machines 
are producing, and almost everyone has found meaning and joy in their lives without 
jobs. There is no enforcement of a rigid equality of personal outcomes across the lives 
of everyone in this society, but there is also no increasingly entrenched divergence 
between those with access to all the latest technologies and those without. 

In chapter 5.5, we confronted the possibility that this society has felt obliged to abandon 
our powerful attachment to the concept of private property, and has moved to some form 
of collective ownership of the means of production, exchange and distribution. In other 
words, socialism. 

Those of us who are convinced that the free market economy, suitably regulated, is an 
ingenious system that has demonstrably created the best standards of living that humans 
have ever enjoyed will find this hard to swallow. Certainly, the idea takes some getting 
used to. 

Advocates of capitalism who have thought this far ahead have suggested that their 
favourite economic model is the best system for an economy of scarcity, but that perhaps 
it won’t be appropriate for an economy of abundance. The problem highlighted by the 
scenario of “the gods and the useless” reveals a different problem. Scarcity hasn’t 
disappeared: it has changed, and become more dangerous. The new scarcity is the 
privileged access to an accelerating flow of powerful new enhancement technologies. 
The danger is that the elite which enjoys this privileged access will rapidly become a 
separate species - a dominant species. Both history and common sense suggest that is 
unlikely to work out well for everyone else. The new scarce resource - the privileged 
access to the cascade of new technologies - is even more valuable and powerful than 
any scarce resource that we value today. 

When normal people read about the lives of billionaires and movie stars, we often think 

they live in a different world. But the distance which separates them from us is tiny 
compared to the gulf which could open up between the Al-owning elite and the 
unemployed majority in a world which passed through the economic singularity while 
retaining private property. 

In chapter 4,1 described a timeline for a successful navigation of the economic 
singularity. It ended up with the Al-owning elites transferring their assets into 
collective ownership, and being hailed as heroes and heroines for doing so. How might 
this work in practice? 


Part of the genius of the market economy - the reason why it is so effective - is that 
decisions are taken by the people best qualified to take them The market enables 
(indeed obliges) each of us to provide truthful signals about what we do and don’t want, 
what we do and don’t value. We buy this car and not that car because we prefer it 
(given our budgetary constraints) and there is no doubt that we are providing a correct 
signal because we are spending our own money. The decisions taken by the market 
overall about how many of each car to make are the aggregate of all these signals. 

By contrast, when decisions are made in a centralised, planned economy, somebody is 
guessing about what is wanted and needed at every level below them. However good 
their data collection system, and however well-intentioned they are, they will always be 
out of date and they will often be just plain wrong. There is also a very good chance 
that corruption will set in, because it is so easy for that to happen. With apologies to 
Lord Acton, [cccxlviii] power corrupts, absolute power corrupts absolutely, and 
corruption is central to centralised planning. 

Common ownership can work well in small communities, such as families, tribes, and 
small villages. But as soon as a society attains any level of size and sophistication, the 
bonds of kinship weaken and individuals start to claim ownership over land and 
property. The society becomes regulated by power structures which begin as means of 
self-defence and evolve into expressions of ambition. 

If (and it is a big “if’) surviving the economic singularity and avoiding fracture means 
ending the system of private ownership, how can this be done without falling into the 
unwelcome embrace of an over-mighty state and centralised planning? 

The answer just might be the blockchain. 


People have gone mad trying to understand how the blockchain works, never mind 
trying to explain it. Its most famous application is Bitcoin, the world’s first completely 
decentralized digital currency, [cccxlix] In just a few years, the Bitcoin “economy” has 
grown larger than the economies of some countries. The value of a Bitcoin has 
fluctuated wildly, hitting a peak of $ 1,216 in November 2013. 

The insights which made Bitcoin possible were published in 2008 under the pseudonym 
Satoshi Nakamoto, and the blockchain is at the heart of it. The blockchain is a public 
ledger which records transactions. The clever bit is that the ledger is completely 
trustworthy despite having no central authority, like a bank, to validate it. It is 
trustworthy in that you can have full confidence that if someone gives you a Bitcoin, then 
you do own that Bitcoin: the person who gave it to you will not be nipping off to spend 
the same piece of currency elsewhere, even though it is entirely digital. 

This confidence arises because transactions are recorded in blocks which are added to 
the chain by people (or rather computer algorithms) called miners. These miners are 
working continuously on mathematical problems whose solutions are hard to find but 
easy to verify. A problem is solved (“mined”) roughly every ten minutes, and each 
solution creates a block. The new block is added to the chain, and incorporates the 
transactions made since the last block was added to the chain. Your transaction is 
published on the blockchain’s network as soon as it is agreed, but it is only confirmed, 
and hence reliable, when a miner has incorporated it into a block. 

Satoshi Nakamoto’s innovation solved a previously intractable challenge in computer 
science known as the Byzantine General’s Problem. Imagine a mediaeval city 
surrounded by a dozen armies, each led by a powerful general. If the armies mount a 
co-ordinated attack, their victory is assured, but they can only communicate by 
messengers on horseback who visit the generals one by one, and some of the generals 
are untrustworthy. The blockchain provides a way for each general to know that a 
message calling for an attack at a particular time is genuine, and has not been fabricated 
by a dishonest general before it reached him. [cccl] 

Digital currency is only one of the possible applications of blockchain technology. It 
can register and validate all sorts of transactions and relationships. For instance, it 
could be used to manage the sale, lease or hire of a car. When you take possession of a 
car, it could be tagged with a cryptographic signature, which would mean that you are 
the only person who could open and start the car. [cccli] 

The revolutionary benefit of the blockchain is that all kinds of agreements can be 
validated without setting up a centralised institution to do so. By removing the need for 
a central intermediary, the blockchain can reduce transaction costs, and it can enhance 
privacy: no government agents need have access to your data without your permission. 

Most importantly, for our present purposes, the blockchain may make possible the 
decentralised ownership and management of collective assets. 

Collective ownership 

Imagine a future in which it is apparent to many people that we are heading towards the 
scenario of the gods and the useless. The elite few who own the machines are as 
uncomfortable as the rest of us about this - or at least a sizeable number of them are. 
They do not want to hand their assets over to a government organisation, as they believe 
this would simply swap one potentially dangerous elite for another one. 

But they realise that if the scenarios of the gods and the useless becomes reality, they 
will end up as pariahs, feared and perhaps hated by the rest of their species. This 
outcome might well be grim for the useless, but it would be unpleasant for the gods as 

I don’t believe the common meme that rich people are all bad, greedy and selfish. I 
have known quite a few, and worked for some of them They seem to me to be the same 
mix of good and bad, greedy and generous as the rest of us. They tend to be smart and 
hard-working, but otherwise they are pretty normal, which is to say, that curious human 
blend of similar and different, happy and sad, predictable and unpredictable. 

When machines do all the paid jobs they will be enormously efficient. Many goods and 
services will be effectively free. Others will be cheap and getting cheaper, but new 
products and services will be being invented all the time, and those which require 
considerable resources will be more expensive. Everyone will have a certain amount 
of UBI, and the way they spend it will give the machines a set of pricing signals to 
enable them to allocate resources sensibly. Life should be good and getting better for 
most people. Protopia could be within reach. 

It strikes me as entirely plausible that in this scenario, the smallish minority which owns 
most of the assets when the game of musical chairs stopped - including notably the AI - 
would prefer to throw in their lot with the rest of us rather than hide behind heavily 
fortified gates, outcasts from the rest of their species. 

It would be a non-trivial project to work out in detail how the assets could be 
transferred into universal common ownership, validated by the blockchain, and 
managed in a decentralised fashion. And it is certainly not a forgone conclusion that the 
rich minority would endorse it. But I suspect it will turn out to be our best way 

Chapter 7. Summary and recommendations 

7.1- The argument 
Automation and unemployment 

This book has argued that improvements in machine intelligence over the next few 
decades are going to make it impossible for most humans to earn a living. It concluded 
that we would be wise to devote some resource to working out how to deal with this 
development - indeed that we would be foolish not to. 

In chapter 2 we looked back at the industrial revolution, and saw how concerns that 
automation would lead to permanent mass unemployment turned out to be unfounded. 
(The Engels pause was lengthy, but not permanent.) Instead, automation raised 
productivity and output across the economy. The unfounded concerns became known as 
the Luddite fallacy. 

In chapter 3.1 and 3.2 we reviewed claims that this time is different. In the information 
revolution, mankind’s third great wave of transformation, machines are increasingly 
able to out-perform humans in cognitive tasks. This might put humans in the 
predicament that horses were placed in by the industrial revolution. 

In 1900, 40% of American workers were employed in agriculture, and that has now 
fallen below 3%. The farmworkers found better jobs elsewhere in the economy, 
sometimes in occupations which their parents could not have imagined. But horses 
didn’t. 1900 was “peak horse” in America, with about 25m of them working on farms; 
now there are fewer than 3m. The difference between the horses and the humans is that 
when machines took over the muscle jobs, humans had something else to offer: our 
cognitive, emotional and social abilities. Horses had nothing else to offer, and their 
population collapsed. 

In chapter 3.3 we heard the response that this is merely a revival of the Luddite fallacy, 
and the rest of chapter 3 explored this debate in detail. 

We considered whether technological automation is biting yet, and saw that the US 
economy has added a significant number of jobs since the banking crisis, although 

incomes have stagnated-. It is hard to disentangle the effects of technology on the job 
market from the effects of globalisation and other effects, but it does not seem likely that 
machine intelligence is creating widespread unemployment yet. 

In chapter 3.4 we reviewed the state of the art in artificial intelligence, and in chapter 
3.5 the exponential rate at which it is improving. We saw that machines are still a long 
way from becoming artificial general intelligences (AGIs) - computers with all the 
cognitive abilities of an adult human, including our flexibility. Even today’s smartest 
machines are still artificial narrow intelligences: they maybe superhuman at playing 
video games, or Go, but they cannot tie a pair of shoelaces or sell you a house, and they 
are not even aware that they are playing a game. 

The fact is that machines don’t need to become AGIs to displace most of us from our 
jobs. They simply have to become better than us at what we do for a living. Because 
they are overtaking us at many forms of pattern recognition, including image recognition, 
speech recognition, and natural language processing, they are in the process of 
overtaking us. And of course, once a machine can do your job, it will quickly be able to 
do it faster, better and cheaper than you can. Machines don’t eat, sleep, get drunk, tired 
or cranky. And unlike human brains, their abilities continue to improve at an 
exponential rate. 

In chapter 3.6 we asked what capabilities people bring to the workplace, and in chapter 
3.7 we reviewed the related technologies which are being introduced alongside AI, 
including personal digital assistants, robotics, virtual reality, augmented reality, and the 
internet of things. We discussed the concerns which these technologies raise about 
privacy, security, isolation and inequality. 

In chapters 3.8 and 3.9 we discussed how widespread unemployment might arise in 
practice, first among drivers, as self-driving vehicles take over their function, and then 
in a range of other industries, including occupations like retail and sales, and the 

In chapter 3.10 we assessed the claim that fears about unemployment are overdone 
because we will invent new kinds of jobs as machines take over the old ones. We 
reviewed the idea that we can race with the machines instead of against them, becoming 
“centaurs”, and occupying ourselves with the “icebergs” of new work which machines 
have made possible. 

But at the end of it all, we concluded that these are only likely to be temporary respites. 
Whatever jobs we invent, the machines will take over most of them as well. In the 

medium term, a large minority of people - perhaps the majority - will not be able to 
earn a living through work. 

The upside 

In chapter 3.11 we saw how this does not have to be bad news. In fact in can be 
extremely good news. Some people are lucky enough to love their jobs, and find 
fulfilment in them For many more people, work is simply a way to generate an 
income. It may provide a purpose, but it does not provide meaning. A world in which 
machines do all the boring work could be wonderful. They could be so efficient that 
goods and services could be plentiful, and in many cases free. Humans could get on 
with the important business of playing, relaxing, socialising, learning and exploring. 
Surely this is what we should be aiming for. 

Chapter 4 provided a timeline - emphatically not a forecast - of a scenario in which 
humanity makes a successful transition towards this enormously positive outcome. 

Challenges and scenarios 

Of course there are challenges, and we turned to these in chapter 5, looking in turn at 
concerns about economic contraction, distribution, meaning, allocation, and cohesion. 

In chapter 5.2 we explored the idea of universal basic income (UBI), a payment made to 
all citizens which allows them to live fulfilling lives when they are no longer able to 
find paid jobs. We heard that thoughtful people in America are concerned that the 
traditional antipathy to socialism will prevent the introduction of UBI, but we concluded 
that once it becomes obvious that plenty of sensible, diligent and hard-working people 
can no longer afford to keep themselves and their families, any such opposition will 
quickly fade. Nevertheless, the fact that people are concerned does underscore the need 
for a public debate about what is coming towards us. 

In chapter 5.3 we discussed the fear that unemployed humans will find their new lives 
hollow, lacking in meaning, and perhaps even boring. We concluded that this too is 
unfounded. For centuries, aristocrats in most countries didn’t work for a living, and in 
many societies they viewed work as a demeaning activity, to be avoided by “people of 
quality”. Some of them got into trouble with drink, drugs and gambling, but only a small 
minority. Most of them seem to have led contented lives, however questionable we 
might find the economic systems they operated in. 

Likewise, retirement is rarely considered a disaster in developed countries. Even 

though most of us only get to enjoy it when we are past our prime, most retirees find 
enough projects and pastimes to keep themselves busy and at peace. Numerous surveys 
have found that happiness is U-shaped: we are at our most content during childhood and 
retirement, and it is probably no coincidence that these are the periods in our lives when 
we don’t work for a living. If we retired when still in our primes, we would be even 
better equipped to enjoy our lives of leisure. 

In chapter 5.4 we considered whether virtual reality might help resolve the problem of 
how to allocate rare goods and services in a world where incomes are hard to vary. 

In chapter 5.5 we tackled what may turn out to be the biggest challenge raised by the 
economic singularity: cohesion. We asked whether capitalism, and in particular the 
institution of private property, will be as suitable for the post-work world as it has been 
during the industrial revolution. This an uncomfortable discussion for people like me 
who believe that a sensibly-regulated market economy has been enormously beneficial 
for humanity. Along with the Enlightenment and the consequent scientific revolution, 
capitalism has made our time the best era to be born human, bar none. 

But a world where almost no-one works, and an elite owns the intelligent machines, is 
going to be a world of fantastic and entrenched inequality. Inequality is often over¬ 
estimated as a contemporary social evil, but this post-economic singularity world will 
also be one where advancing technology makes available radical enhancements to our 
physical and cognitive performance. These enhancements will come along faster and 
faster, and groups with privileged access to them may start to diverge from everyone 
else, and become a separate species. The author Yuval Harari has referred to this 
scenario in the chilling phrase, “the gods and the useless”. The “Brave New World” 
depicted by Aldous Huxley in 1931 might be one of the least bad outcomes of this 

If the post-economic singularity world needs a different type of economy, then we need 
to start thinking now about what that might be - and also how to get there. The damage 
that could be caused by an uneven or violent transition to the new world could be 

In chapter 6 we pulled these threads together with a review of half a dozen potential 
scenarios. Chapter 6.1 presented and dismissed the idea that technological progress has 
slowed almost to a halt, and there is nothing to worry about. Chapter 6.2 rehearsed the 
hope that we can race with the machines by becoming centaurs and enjoying the 
icebergs of new work. Chapter 6.3 offered the idea that unemployment will grow, but 
can be accommodated by UBI. 

Chapter 6.4 reprised the scenario of the gods and the useless, and chapter 6.5 reminded 
us that civilisation is fragile, and that a poorly-planned transition towards a new 
economy could be hazardous. Chapter 6.6 adopted Kevin Kelly’s term Protopia for a 
successful transition, and suggested that the blockchain might turn out to be the 
mechanism to administer society’s collectively owned assets, notably its artificial 

7.2 - The two singularities 

In my previous book, “Surviving AI”, I wrote at length about the challenge and the 
opportunity presented by the technological singularity, the moment when (and if) we 
create an artificial general intelligence which continues to improve its cognitive 
performance and becomes a superintelligence. Ensuring that we survive that event is, I 
believe, the single most important task facing the next generation or two of humans - 
along with making sure we don’t blow ourselves up with nuclear weapons, or unleash a 
pathogen which kills everyone. 

If we secure the good outcome to the technological singularity, the future of humanity is 
glorious almost beyond imagination. As DeepMind co-founder Demis Hassabis likes to 
say, humanity’s plan for the future should consist of two steps: first, solve artificial 
general intelligence, and second, use that to solve everything else. “Everything else” 
includes poverty, illness, war and even death itself. 

The stakes in the economic singularity are not so high (which is why I tackled it 
second.) If we find ourselves in the “gods and the useless” scenario, or if our societies 
collapse as we fail to transition from modern capitalism to something more suitable for 
the new world, it is unlikely that every human will die. (Not impossible, though, as 
someone might initiate a catastrophic nuclear war.) Civilisation would presumably 
regress, perhaps drastically, but our species would survive to try again. Trying again is 
something we are good at. 

On the other hand, if it is coming at all, the economic singularity is coming sooner than 
the technological singularity. No-one knows how long it will take to build an artificial 
general intelligence, but it looks tremendously hard. It is probably only a matter of time, 
but that time may well be quite a few decades. The economic singularity is likely to be 
with us in two or three decades - perhaps not in the sense that a majority of people will 
be unemployable by then, but in the sense that it will be obvious and undeniable that it is 
going to happen. Asset prices may collapse at that point. 

Relatively speaking, then, the technological singularity is more important but less 
urgent, while the economic singularity is less important but more urgent. 

7.3 - What is to be done? 

Relinquishment won’t work 

Impressed by the dangers attending the two singularities, you might think it would be a 
good idea to call a halt to further development of artificial intelligence research, either 
permanently, or simply for long enough to allow us to work out how to ensure that both 
events are beneficial. Unfortunately this is impossible. 

First, we would not know what research to pause. Improvements in the performance of 
AI comes from many directions: chip design and manufacture, algorithm development, 
the accumulation and statistical analysis of data, to name but three. Unless we could 
arrest pretty much all scientific and technological research, we could not be sure that 
someone, somewhere, was not working on something which will advance AI. 

Second, the incentive to develop and deploy a better AI than the competition is literally 
irresistible. For companies like Google and Facebook, who are leading the way in AI 
research, it is a matter of critical commercial performance in the short term, and of 
economic survival in the medium and long term. For military commanders it is quite 
literally a matter of life and death. Even if by some miracle all the world’s leading 
politicians could be gathered together and persuaded to sign a joint declaration that all 
AI research will stop, they would not abide by it. We can all agree that North Korea 
would cheat, but who would be so naive as to think that their own government would 
not do the same? 

If it is possible to create an artificial general intelligence, it will be created - and it 
will be created as soon as it becomes possible. The same applies to the technologies 
required to render most humans unemployable. 


There are at least four permanently-established organisations studying the risk to 
humanity posed by the potential arrival of superintelligence. [ccclii] There is only one 
that I know of which is studying the future of automation and technological 
unemployment. [cccliii] There should be more. 

In chapter 4 we explored how hard it is to make accurate forecasts, but failing to keep a 

lookout for approaching dangers (and opportunities) is foolish. The view that most 
people will be rendered unemployable by machine intelligence within the next few 
decades is probably a minority opinion at the moment, but a great many people are 
uncertain about what will happen, and the case argued in this book and elsewhere is 
surely at least plausible enough to be worth watching out for. We should be employing 
economists and others to monitor the available data for signs of technological 
unemployment, and devising new ways to detect it. 

The economist Robin Hanson thinks that machines will eventually render most humans 
unemployed, but that it will not happen for many decades, probably centuries. Despite 
this scepticism, he proposes an interesting way to watch out for the eventuality: 
prediction markets. People make their best estimates when they have some skin in the 
forecasting game. Offering people the opportunity to bet real money on when they see 
their own jobs or other peoples’ jobs being automated may be an effective way to 
improve our forecasting, [cccliv] 


We don’t have sufficient information to draw up detailed plans for the way we would 
like our economies and societies to evolve. But we can, and probably should, be doing 
detailed scenario planning. 

Scenario planning has been practised by military leaders since time immemorial. It was 
given the name by Herman Kahn, who wrote narratives about possible futures for the US 
military while working for the RAND Corporation in the 1950s. (His suggestion that a 
nuclear war might be both winnable and survivable made him one of the inspirations for 
Dr Strangelove in the classic 1964 movie, [ccclv] ) Scenario planning was adopted by 
Shell Oil after it (along with the rest of the industry) was disastrously wrong-footed by 
the rise of the oil cartel OPEC in the 1970s. [ccclvi] 

Scenario planning is more art than science, but it can be a valuable discipline. When 
we commit our thoughts about a possible future to paper we are forced to consider them 
rigorously. Institutes consisting of smart people doing this work could make a valuable 

An informal version of this is the daily business of futurists and futurologists, people 
who are often viewed with scepticism by the wider public. Perhaps that will change - 
in fact, perhaps futurology will come to be seen as a mission-critical profession. 

Science fiction writers also have an important role, in providing vivid metaphors and 


The role of the tech giants 

Google, Facebook, Amazon, Microsoft, IBM and Apple are shaping the new world we 
are moving into, along with their Chinese counterparts Baidu, Alibaba and Tencent. 
Their motivation is partly commercial: they understood sooner than anyone else that 
artificial intelligence and related technologies will increasingly provide most of the 
world’s economic value. They are moving aggressively to dominate the AI space, and 
competing fiercely with each other for talent and market positions. 

Although I have no privileged access, it seems to me that many of the leading figures in 
this industry are also motivated by something else: a belief that the future will be better 
than today, and an impatience to make it arrive faster. 

It is ironic, then, that these companies are often reluctant to talk about their vision. In 
particular, they shy away from discussing AI. It is understandable: every time they talk 
to a journalist about AI, the resulting article is accompanied by a picture of the 
Terminator. It must be enormously frustrating to be working hard to conjure a happier, 
richer, safer world, when all you get in response is talk about existential risks. 

It may be understandable, but it is also dangerous. The idea that artificial intelligence is 
improving quickly is now firmly in the public mind. When self-driving cars become 
common, smartphones are capable of sensible conversations, and domestic robots can 
carry out many of our domestic chores, people will increasingly ask where it is all 
heading. In the absence of optimistic answers, they will gravitate towards the bad ones, 
and Hollywood has given us plenty of those. 

We need potent new memes, illustrating the current benefits and the future promise of 
AI. The tech giants are creating this new world; even if only for their own self- 
preservation, it would be a good idea for them to explain how it is capable of being a 
glorious new world. 

What should I study? 

The question that young people (and their parents) naturally ask about the economic 
singularity is, how can I best prepare for the economy that we are moving towards? It’s 
an important question for me, too: at the time of writing my son is 15. 

The obvious answer is to study computing. Computers are at the heart of the changes 
sweeping the world in the information revolution, so it has to be valuable to understand 
how they work and what they can and cannot do. If possible, study machine learning, 
and in particular, deep learning. It seems a safe bet that these powerful techniques will 
remain important for years to come. Carrying out maintenance, supervision or 
development of hardware of software may keep you in a job for longer than most. 

In the long run, however, if the argument of this book is correct, we are probably all 
unemployed. It may well be an advantage for a while to be rich, but if we manage the 
transition successfully that may become less important and less worthwhile. And if we 
don’t... well, let’s just say we have to. 

Beyond the economic singularity you’re going to want to have as rich an interior life as 
possible, so give yourself as broad an education as you can. Studying your own and 
other people’s languages will give you insights into how our minds work. Studying 
sciences will give you insights into how the world works. And studying the humanities 
will give you insights into how societies work. All of these should make what could be 
a very long life an interesting one. 

The most important generations 

Every generation thinks the challenges it faces are more important than what has gone 
before. They can’t all be correct. American journalist Tom Brokaw bestowed the name 
“the greatest generation” on the people who grew up in the Great Depression and went 
on to fight in the Second World War. As a “baby boomer” myself, I certainly take my 
hat off to that generation. 

Speaking at the United Nations in 1963, John F Kennedy said something which would 
not sound out of place today: "Never before has man had such capacity to control his 
own environment, to end thirst and hunger, to conquer poverty and disease, to banish 
illiteracy and massive human misery. We have the power to make this the best 
generation of mankind in the history of the world - or make it the last." [ccclvii] 

Today’s rising generation is the Millennials, born between the early 1980s and the early 
2000s. They are also known as Generation Y, and the one after them, born from the 
early 2000s to the early 2020s, is provisionally called Generation Z. Let’s hope that is 
not prophetic. 

The Millennials and Generation Z have been born at the best time ever to be a human, in 
terms of life expectancy, health, wealth, access to education information and 

entertainment. They have also been born at the most interesting time, and the most 
important. Whether they like it or not, they have the task of navigating us through the 
economic singularity of mass unemployment, and then the technological singularity of 
super-intelligence. If they succeed, humanity’s future is almost incredibly good. If not, 
it could be bleak. It will fall largely to them to plot the course, adjust it where 
necessary, avoid the rocks and the cries of the Sirens, and bring the ship safely home. 


I am enormously grateful to the following people, who have given their time and energy 
to support this book, and in many cases to provide constructive criticism on its earlier 
drafts. I have learned a lot from their insights, and the book is much better for them. All 
errors, omissions and solecisms are of course my fault, not theirs. 

Adam Jolly, Adam Singer, Aubrey de Grey, Ben Medlock, Ben Goldsmith, Chris 
Meyer, Clive Tinder, David Wood, Gerald Huff, Hugo de Garis, Jeff Pinsker, Jim 
Muttram, Justin Stewart, Kenneth Cukier, Malcolm Myers, Peter Fenton O'Creevy, Peter 
Monk, Randal Koene, Roman Yampolskiy, Stuart Armstrong, Will Gilpin, William 
Charlwood, William Graham. 

As always, my profoundest thanks go to my partner Julia, who is my adviser, my 
cheerleader, and my kindest but most penetrating critic. My hugely talented designer 
Rachel Lawston has again produced a cover which (I think) looks great and answers the 
brief perfectly. 

m Memes are ideas or beliefs which spread from person to person to become pervasive within a culture. 

an Many textbooks place the start of the industrial revolution in the second half of the 18 th century, but I 

like the argument that Thomas Newcomen's creation of the first practical steam engine in 1712 provides the best 
origin story. 

liiil There is no general agreement about when the information revolution started. In his 1962 book 

“The Production and Distribution of Knowledge in the United States”, the Austrian economist Fritz Machler 
suggested that with 29% of GDP accounted for by the knowledge industry, it had begun. 

m The term was first applied to human affairs back in the 1950s by John von Neumann, a key figure in 

the development of the computer. The physicist and science fiction author Vemor Vinge argued in 1993 that 
artificial intelligence and other technologies would cause a singularity in human affairs within 30 years. This idea 
was picked up and popularised by the inventor and futurist Ray Kurzweil, who believes that computers will 
overtake humans in general intelligence in 1929, and a singularity will arrive in 2045. singularity 

M The event horizon of a black hole is the point beyond which events cannot affect an outside observer, 

or in other words, the point of no return. The gravitational pull has become so great as to make escape impossible, 
even for light. 

mj _ 

utm content=buffer71a7e&utm medium=social&utm campaign=buffer 


At the end of this video: lntp:// 




W _ 

Ml 1 .pdf. Employment in agriculture declined in 

absolute terms as well, from 11.7m in 1900 to 6.0m in 1960. 

MU _ -census-analysis/ 170-years-of- 


Mill http://www.americanequestrian.corn/pdf/us-equine-demographics.pdf 

[xiv] https ://en. wikipe dia. org/wiki/ Automation#c ite_note -7 

[xv] M. A. Laughton, D. J. Warne (ed), Electrical Engineer's Reference book 

MU _ 

invented/artic Ie_c7a00da2-b3eb-54el-9c8d-ee36483a7e33.html 

[xvil] Radio frequency Identification tags. They can take various form s - for instance, some have 

inbuilt power sources, while others are powered by interacting with nearby magnetic fields, or the radio waves 
which interrogate them. 

[xvill] _ 


MX] _ http://www.eater.eom/2016/5/5/11597270/kfc-robots-china-shanghai 

Md] _ http://www.ehow.coiu/about_4678910_robots-car-mannfactnring.html 






httpf/jetpre ss. org/v24/ c ampa2. htm 

[xXYl] Ricardo originally thought that innovation benefited everyone, but he was persuaded by Malthus that 

it could suppress wages and cause long-term unemployment. He added a chapter called “On Machinery” to the final 
edition of his book “On the Principles of Political Economy and Taxation”. 



[xxviii] law 


pause-a-pessimist-s-guide-to-the-british-indus trial-revolution 


This depends on the two planets being pretty much as close as they ever get. 


[xxxill] This and the other quotes in this paragraph and the next one are from Chapter 10: 

Toward a New Economic Paradigm. 

[xxxiv] Brynjolfsson is the director of the MIT Center for Digital Business 

and McAfee is a principal research scientist there. 

[xxxv] The word “inequality” crops up 42 times in the book, including in the 

titles of sources, but the authors never explicitly connect it with “spread”. 

[xxxvi] The loosely-organised protest organisation that sprang up after the 2008 credit crunch to 

campaign against inequality. 

[xxxvii] Chapter 12: Learning to Race with the Machines: Recommendations 

for Individuals. 


Chapter 13: Policy Recommendations. 




Chapter 14: Long-Term Recommendations. 






_ https://www.reddit.eom/r/BasicIncome/ and https://www.reddit.eom/r/basicincome/wiki/index 

[xliii] _ . 


https: // www .youtube, com/ watch? v=C5MVXdg6riho . 




[xlvii] https ://newrepublic, c om/ artic le/69326/ c all-the - wolf 

[xlviii] _ http://www.ft.eom/cms/s/0/dfe218d6-9038-lle3-a776-00144feab7de.html#axzz3stkJblV2 

[xlix] The Programme was established in January 2015 with funding from 

Citibank, one of the largest financial institutions in the world. The Oxford Martin 
school was set up as part of Oxford University in 2005, as an institution dedicated to 
understanding the threats and opportunities facing humanity in the 21 st century. It is 
named after James Martin, a writer, consultant and entrepreneur, who founded the 
school with the largest donation ever made to the university - which was no mean 
feat given that Oxford was founded 1,000 years ago, and is the oldest university in 
the world (after Bologna in Italy). 

[1] _ 

robots-bv-2025. html 

Hi] _ 

Koplg. l&utm 

Oil] _ 


liiii] _ technology/four fundamentals of workplace automation 

m _ 

wont- go-to-the -moon/ 

IlY] _ https://www.aeaweb.Org/articles.php7doFl0.1257/iep.29.3.3 

IlYi] _ littps//reason, com/archive s/20 1 5/03/03/how-to-survive-a-robot-uprisin 

liYii] _ http;//www.p() 13/11 /the-robots-are-here-098995 

[lviii] http: / / w w w. forbes .com/ sites/danschawbel/2015/08/04/geoff-col vin- 



[lx] German academic Marcus Hutter, and Shane Legg, co-founder of 


m _ http://www. 

[lxii] The Shape of Automation for Men and Management by Herbert Simon, 



Computation: Finite and Infinite Machines by Marvin Minsky, 1967 




http://www. etvmon line, com/index. php?term=a1gorithm 2016 / 01 /microsoft-neural-net-shows-deep-leaming-can-get-wav-deeper/ 

[lxvii] Moravec wrote about this phenomenon in his 1988 book “Mind Children”. A possible 

explanation is that the sensory motor skills and spatial awareness that we develop as children are the product of 
millions of years of evolution. Rational thought is something we have only been doing for a few thousand years. 
Perhaps it really isn’t hard, but just seems hard because we are not yet optimised for it. 

[lxviii] https :// 


lixx] _ 







[lxxiv] _ 





_ full-autonomous-dr iving-from-3-vears- 


[lxxvii] _ Google Is God.html 


https:// www. re ddit. c om/ r/churchofgoogle/ 

The answer, if you're searching from England, is to fly over Asia. 



utm campaign=socialflow&utm source=facebook&utm medium=social 

[lxxxi] _ 

[lxxxii] _ http://www.thedrum.eom/opinion/2016/02/08/why-artificial-intelligence-kev-google-s-battle- 


[lxxxiii] _ 

[lxxxiv] They are the Pembroke and the Cardigan Corgi. http://research.rmcr0S0ft.C0m/en- 


[lxxxv] _ http://image-net.Org/challenges/LSVRC/2015/index#news 


httpy/,asp?doc id=1325712 


httpsy/ WgclJOsBk?t=33 

[lxxxviii] _ 



_ httpy/www. 


M _ httpy/www. 

jxci] At the time of writing, April 2016, Aipoly is impressive, but far from perfect. 

[xcil] _ httpy/ 



_ httpy/www. 


[xciv] _ 1 /baidus-deep-lcarning-svstcm-rivals-pcoplc-at- 


[xcv] _ https :// 

[xcvi] _ 


[xCVlil] http://www. 



LcJ _ 



2013 data: 
development/2013/stb-gerd-2013 .html 

Mi] _ 

[ciii] intelligence/ 

[civ] _ 

M] _ 


[CYi] _ 

[evil] The name Parsey McParseFace is a play on a jokey name for a research ship which received a 

lot of votes in a poll run by the British government in April 2016. 

[cvill] Assuming you don't count the Vatican as a proper country, 




tel _ 

teJ _ 

[cxil] Haitz's Law states that the cost per unit of useful light emitted decreases exponentially 

[cxiii] cpo decreasing, php 


[cxv] . 



. The “TTT-V” refers to the periodic table group the material belongs to. Transistors made from these 
semiconductors should consume far less power, and also switch much faster. 



[cxvill] _ http://www. 

[cxix] _ 

[exx] Clock speed, also known as clock rate or processor speed, is the number of cycles a chip (central 

processing unit, or CPU) performs each second. Inside each chip is a small quartz crystal which vibrates, or 
oscillates, at a particular frequency. It takes a fixed number of oscillations, or cycles, to perform the instructions 
that a chip is given. One cycles is one Hertz, and today’s chips operate in gigaHertz (GHz), billions of cycles per 
second. As other aspects of chip designs diverge, clock speed is no longer a reliable measure of a chip’s effective 

[CXXI] _ 


[cxxil] _ 




[CXXV] CPU stands for Central Processing Unit. They are general purpose processors which can carry 

out many kinds of computation, but are not necessarily optimised for any of them. GPU stands for Graphics 
Processing Unit, and as the name suggests, they were originally designed for displaying graphics in video games. 
They are very good at taking huge quantities of data and carrying out the same operation over and over again. It 
turns out that machine learning benefits from their particular capabilities. CPUs and GPUs are often deployed in 

[CXXVI] _ 



[CXX1X] _ 

[CXXX] 2013 data: 315661.pdf 

[CXXXI] This has given rise to the term “subtractive manufacturing” for the traditional for ms of 

manufacturing. This method of naming is rather splendidly called a retronym. 


[cxxxiii] _ http://www. rfidioumaL c om/ artic le s/view?4986 




Coined by another British entrepreneur, Simon Birrell: 

[cxxxvi] _ 






[CXXXIX] http://www.intbrmation-age.eom/it-management/strategv-and-innovation/l 23460379/trains- 


Ml] _ 



[cxlii] Not an everyday object outside the USA, of course 

[cxliii] _ 


[cxliv] This is actually a great idea, which is being trialled in Argentina at the time of writing: 

safetv-screens-set-to-save-overtaking-drivers.html Of course it may be less valuable when cars drive themselves 
and their human occupants don’t look at the road. 

[cxlv] An ocarina is a wind instrument about the size of a fist. First introduced to Europeans by the 

Aztecs, it looks like a toy submarine. 

[cxlvi] https :// www. youtube. c om/watc h? v=OQ N iZfSs P c 0 

[cxlvii] _ 

[cxlviii] _ 


m _ 

Mi] _ 

Mill 1 6/05/robots-are-starting-learn-touch/ 1 28065/ 



[cliv] https://www.theguardian.corn/technobgv/2015/aug/03/hitchbot-hitchhiking-robot-destroved- 


[civ] _ 


[clvi] http://techcrunch.eom/2016/01/07/the-grillbot-is-a-robot-that-cleans-vour-grill/#.w9z87m:Hd0d 

[clvii] httpy/ 


[clviii] _ httpy/intl, eksobionics. com/ 

[clix] Your brain is wired to make you see things before you hear them as it knows that light travels 

faster than sound. Thus the brain can tolerate audio lagging video, but is much less tolerant of video lagging audio. 
This is known as Multi-Modal Perception. 

[Clx] _ http://www.ft.eom/cms/s/0/b33d75fe-cc5a-lle5-be0b-b7ece4e953a0.html#axzz3znOxP8QH 

[clxi] httpy/www.kit 



_ httpy/www. digi-capitaLcom/news/2015/04/augmentedvirtual-realitv-to-hit- 150-billion-disrupting-mobile - 

bv-2020/#. V oV 65 vmLRD 8 



[clxiv] _ 


[clxv] The games industry is much bigger than Hollywood if you stop measuring move income at the 

box office. If you add in DVD and other “windows”, plus merchandising, it is hard to say. 

[clxvi] httpsy/versions. 

[clxvii] _ 

httpy/ crump and harwood%2C the net closes around us/ 

[clxviii] _ httpsy/www. 

calculating-vour-threat-score/2016/01/10/e42bccac-8e 15-Ile5-baf4-bdf37355da0c_storv.html 





[clxxii] _ 



[clxxiv] _ 




utm source=Twitter&utm medium=tweet&utm campaign=@KvleSGihson 
[clxxvi] The Flynn Effect: 

[clxxvii] WHO "Global Status Report on Road Safety 2013: supporting a decade of action 

[clxxviii] _ 


[clxxix] _ 


[clxxx] _ 

[clxxxi] _ http://www.etvmonline. com/index.php?term=autocar 

[clxxxii] _ 

[clxxxiii] I NHTSA/Press I Releases/U.S. I Department I of I Transportation l Releases I Policy I on 


[clxxxv] _ 

[clxxxvi] _ 


[clxxxvii] _ 

[clxxxviii] _ 


[clxxxix] _ 





_ http://venturebeat.eom/2016/01/10/elon-musk-youll-be-able-to-summon-vour-tesla-from-anvwhere-in- 



_ https://www.washingtonpost.eom/news/the-switch/wp/2016/01/ll/elon-musk-savs-teslas-autopilot-is- 



[cxciv] _ 




[cxCVll] _ 


[CXCIX] It has been suggested that electric cars should make noises so that people don’t step off the 

pavement in front of them. A friend tells me he would like his to make a noise like two coconuts being banged 
together, in homage to the scene in Monty Python and the Holy Grail where King Arthur, unable to afford a horse, 
has a camp follower fake the noise of one with coconuts. 

[cc] _ http://www.pcmag.eom/article2/0.28l7.2370598.00.asp 

[cci] _ 

research-center.him 1? r=0 




[cciv] _ 


[cCVl] _ httpy/www. asp 

[cCVll] . 




. http://www.bls,gov/ooh/transportation-and-material-moving/taxi-drivers-and-chauffeurs.htm 

[ccx] See chapter 3.10. 

[cCXl] _ 

pricing-12-18-percent 20150325.html 

[ccxii] The Economist , December 4, 2003 






[cCXVl] _ 

[ccxvii] _ https://en.wikipedia.Org/wiki/Horn_%26_Hardart#Automated_ food 



[CCXIX] _ hoar/15-04-14-death of a b2b salesman 

[ccxx] _ 



[cCXXll] https://www.technologvreview.eom/s/601215/china-is-building-a-robot-armv-of-model- 


[cCXXlll] _ 



[CCXXV] _ https :/7www. youtube. coni/watclV?v=11 XKDiiqMOUhv 

[ccxxvi] 22794242.htm 

[ccxxvii] _ 

[ccxxviii] _ 



[CCXXX] _ http://www.le 

unloc king-innovation-in-vour-law-firm 







[cCXXXVll] _ 

[cCXXXVlll] _ 

than-humans/?utm content=bufferb9e5d&utm medium=social&utm campaign=buffer 

[CCXXXIX] _ http://forbcsindia.corn/article/hidden-geins/thvrocare-technologies-testing-new-waters-in- 


[ccxl] _ 


[ccxli] _ 

[ccxlii] _ http://www.ft.eom/cms/s/2/dced8150-b300-11 c5-8358-9a82b43f6b2f.html#axzz3xL3RoRdv 


_ http://www.ft.eom/cms/s/2/dced8150-b300-l le5-8358- 


[eexliv] http://www.theverge.eom/2016/3/10/l 1192774/demis-hassabis-interview-alphago-google- 


[ccxlv] _ 


[ccxlvi] _ 



[ccxlviii] _ 




[ccli] https ://www. 


[cclii] _ 


[ccliii] _ https://www.sigfig.eom/site/#/home 


alLhtml? r=0 



[cclvi] _ 

[cclvii] _ (Paywall) 

[cclviii] _ http://www.ft.eom/cms/s/0/5eb91614-bee5-lle5-846f-79b0e3d20eaf.html#axzz3zEmSvuZs 

[cclix] _ 



[cclxi] _ 




[cclxiii] _ 


[cclxiv] Assuming the work is happening on Earth. Wikipedia offers a more general but less 

euphonious definition: “Work is the product of the force applied and the displacement of the point where the force 
is applied in the direction of the force.” 

[cclxv] _ 

[cclxvi] kellv-the-techninm 


[cclxviii] _ 


[cclxix] _ 

[cclxx] 5/03/a-brie f-historv-of-the-atm/388547/ 

[cclxxi] _ 

concerns#stream/ 0 

[cclxxiii] _ httpy/viterbi. therapist.htm 

[cclxxiv] _ 


[cclxxv] _ 

[cclxxvi] _ http://www. 


[cclxxviii] _ 

[cclxxix] _ http://www.ft.eom/cms/s/2/c5cf07c4-bf8e-lle5-846f-79b0e3d20eaf.html#ax7z3vLGlrrlJ 

[cclxxx] _ http://www.bls. gov/cps/cpsaat 11 .htm 

[cclxxxi] Man%27s Sky 



[cclxxxiii] http7/www. famous-entrepreneurs.html 

[cclxxxiv] _ 

[cclxxxv] _ 

[cclxxxvi] 1 5 

[cclxxxvii] _ 

[cclxxxviii] _ 


[cclxxxix] _ http://www.ft.eom/cms/s/0/dfe2 1 8d6-9038- 11 c3-a776- 


[CCXC] _ 

[eexei] As noted in chapter 3.4, Anders Sandberg is James Martin Fellow at 

the Future of Humanity Institute at Oxford University. He was referring to Elon 
Musk's warning that we might be the boot loader for a digital superintelligence, 
meaning that we create it and then disappear. Anders suggests that a better fate 
would be what happened to a prokaryotic cell which was absorbed by another, larger 
cell and became an essential component of a combined, more complex entity, the first 
eukaryotic cell. 

[eCXCll] _ 

[eCXCiii] _ 1/11/16/robots-buv-cars/ 





[cCXCVlll] _ 



https://www.reddit.eom/r/BasicIncome/wiki/index#wiki that.27s all very well.2C but where.27s the evidence.: 


[cCCl] _ 


utm_content=buffer71a7e&utm_medium=soc ial&utm_source=plus. google. com&utm_campaign=buffer 

[cCClll] _ 


[CCCIV] _ 51227-story. Iitml 

[CCCV] _ httpy/ 

[CCCVI] _ laws in the United States#References 



[cCCviii] _ n 198196.html 

[CCCIX] _ 

[CCCX] _ 

[cCCXl] _ 

inflation-fe7 ld69f 15e7#. 3vezsngei 


[cCCXlll] _ http://www.forbes.eom/sites/greatspeculations/2012/12/05/how-i-know-higher-taxes-would- 



[CCCXV] _ curve 


[cCCXVll] A min or character in Shakespeare’s Henry VI called Dick the Butcher has the memorable 

line, “First thing we do, let’s kill all the lawyers.” It seems Shakespeare was not fond of lawyers: 

[cCCXVlll] _ 


[CCCXIX] _ average earnings 2014.html 

[CCCXX] _ 


[cCCXXl] _ 


[cCCXXll] _ 


[cCCXXlll] _ 



[CCCXXV] I’m indebted to Dr Justin Stewart, an investor, for prodding me to address the issue of assets 

more closely. 


[cCCXXVll] In case you only recently arrived on this planet, that was a reference to the sainted Douglas 

Adam’s “Hitchhiker’s Guide to the Galaxy” series. If you haven’t read it, I recommend that you put this book 
down and read that one instead. I won’t be offended. But please come back here afterwards. 

[cCCXXviii] _ 

[CCCXXIX] The novel is sometimes said to have originated in the early 18 th century, but in fact it is a 

much older art form. What happened then was that writers began publishing books which described life as they 
actually saw it. https://en.wikipedia. 0 rg/wiki/N 0 vel# 18th century novel 

[CCCXXX] I am indebted to AGI researcher Randal Koene for this observation. 

[cCCXXxi] _ https7/ Davis 

[cCCXXxii] _ 



[cCCXXxiv] See chapter 3.1 


[cCCXXXVl] _ 


[cccxxxvii] _ noah harari what explains the rise of humans/transcript?language=en 

[cCCXXXVlll] _ httpy/motherboard. 


[CCCXXXIX] Covered in detail in my previous book, “Surviving AI”. 

[CCCXI] _ and drugs and rock and roll 

[cccxli] I am that terrible old cliche: a socialist student whose left-wing views did not long survive 

contact with the world of work. As a trainee BBC journalist writing about Central and Eastern Europe long before 
the Berlin Wall fell, I soon realised how fortunate I was to have grown up in the capitalist West. I didn’t expect to 
be heading back in the other direction in later life. 

[cccxlii] _ https J/e markoff-the-next-wave 

[cccxliii] _ httpy/uk. 


[cCCxliv] _ experiment 

[cCCxlv] _ httpy/ 

[cccxlvi] _ httpy/ 

[cCCxlvii] _ https://www. e dge. or g/c onvers ation/ke vin_ke lly-the -te c hnium 


[cccxlix] _ 

[CCCI] _ 

[cccli] _ 

innovation/ analysis/money-no-middleman/ tab/1 

[ccclii] : The Machine Intelligence Research Institute (MIRI) in Northern 

California, The Future of Humanity Institute (FHI) and the Centre for the Study of 
Existential Risk (CSER) in England’s Oxford and Cambridge respectively, and the 
Future of Life Institute (FLI) in Massachusetts. 

[cccliii] The Oxford Martin Programme on Technology and Employment, part 

of Oxford University and part-funded by Citibank. There have been numerous less 
permanent initiatives, such as the 2015-16 research programme of Fundacion Banldnter, 
one of Spain’s leading banks. (Disclosure: I am involved in that one.) 

[eCCliv] _ archive/2003/09/15/349149/index.htm 

[ccclv] Paul Boyer, 'Dr. Strangelove', a chapter in “Past Imperfect: History According to the Movies” 

edited by Mark C. Carnes 

[ccclvi] _ 


[eCClvii] _